diff --git a/README.md b/README.md
index 2b8622a1..536e7e6e 100644
--- a/README.md
+++ b/README.md
@@ -838,7 +838,7 @@ The device comes from what the cluster advertises, not from the host alone:
 
 | Cluster has | Device | Pod gets |
 |-------------|--------|----------|
-| allocatable `habana.ai/gaudi` | `HPU` | one Gaudi, served by vLLM (see [vLLM on Gaudi](#vllm-on-gaudi)) |
+| allocatable `habana.ai/gaudi` | `HPU` | one Gaudi, served by vLLM, not yet run on Gaudi hardware (see [vLLM on Gaudi](#vllm-on-gaudi)) |
 | allocatable `gpu.intel.com/xe`, else `gpu.intel.com/i915`, on a node labelled `gpu.intel.com/xmx=true`, skipping a driver labelled `k3d-gpu.io/xmx.<driver>=false` | `GPU` | one GPU, a `gpu.intel.com/xmx: "true"` node selector; the host render group is added as a supplemental group |
 | anything else, including an Intel GPU without XMX (or with `xmx` unknown) | `CPU` | no device; the P-cores on a hybrid node; on a node labelled `...cpu-cpuid.AMXTILE=true` it is pinned there |
 | `npu.intel.com/accel`, only with `K3D_GPU_AGENT_DEVICE=NPU` | `NPU` | the NPU and its accel node's group |
@@ -977,8 +977,8 @@ OpenAI API at `/v1` from `svc/vllm`:
 
 | Backend | Image (pinned by digest) | Notes |
 |---------|--------------------------|-------|
-| `vllm` | `vllm/vllm-openai-xpu:v0.32.0` | upstream vLLM for Intel GPUs (Data Center GPU and Arc, per [vLLM's XPU guide](https://github.com/vllm-project/vllm/blob/v0.32.0/docs/getting_started/installation/gpu.xpu.inc.md)); v0.32.0 not yet run here on an Arc B-series card |
-| `llm-scaler` | `intel/llm-scaler-vllm:1.5` (vLLM 0.26.0-b2) | Intel's build for Arc B60/B70 |
+| `vllm` | `vllm/vllm-openai-xpu:v0.32.0` | upstream vLLM for Intel GPUs (Data Center GPU and Arc, per [vLLM's XPU guide](https://github.com/vllm-project/vllm/blob/v0.32.0/docs/getting_started/installation/gpu.xpu.inc.md)); served end to end here on an Arc Pro B70 at release 0.1.94 (status below) |
+| `llm-scaler` | `intel/llm-scaler-vllm:1.5` (vLLM 0.26.0-b2) | Intel's build for Arc B60/B70; not yet run here on hardware (status below) |
 
 ```bash
 K3D_GPU_AGENT_BACKEND=vllm k3d-gpu agent up
@@ -1000,9 +1000,39 @@ other backend first so its GPU is free; `agent down` and `agent status` cover bo
 The chat UI below and other clients that use `svc/ovms` at `/v3` need
 `http://vllm.k3d-gpu-agent.svc:8000/v1` instead.
 
-Not yet run on Arc hardware here: serving end to end and tensor parallelism
-above 1 (`CCL_ZE_IPC_EXCHANGE=pidfd`; oneCCL calls `prctl(PR_SET_PTRACER, getppid())`,
-so sibling workers should need no `CAP_SYS_PTRACE` under yama `ptrace_scope=1`).
+**Status: `vllm` is verified end to end on an Arc Pro B70 at release 0.1.94.** On
+2026-11-08, release 0.1.94, installed by the public `packaging/install.sh`, ran
+`k3d-gpu up` and `K3D_GPU_AGENT_BACKEND=vllm k3d-gpu agent up` with the defaults and no
+patching: `vllm/vllm-openai-xpu:v0.32.0` (vLLM 0.32.0, digest `sha256:20509fab...`)
+serving `Qwen/Qwen3-4B` at revision `1cfa9a72`. The pod was Ready on one card
+(`gpu.intel.com/xe: 1`) and the launcher's warm-up was answered. Through the
+port-forward `agent up` prints, "What is 17 times 23?" came back 391, and the `hermes`
+parser returned the tool call. The host was a Xeon 6741P with 4x Arc Pro B70 (`xe`,
+kernel 7.0.0-31, Ubuntu 26.04.1). The same checks passed in the same hold on the
+`v0.33.0` bump's head ([#357](https://github.com/88plug/k3d-gpu/pull/357)). Record, with
+every version and file:
+[bench/vllm-agent/results/2026-11-08](bench/vllm-agent/results/2026-11-08/README.md).
+
+Releases 0.1.21 (2026-10-16) through 0.1.93 do not serve with vLLM, on Arc or on a
+Gaudi, and neither does `agent up --model` when the fit picks vLLM. The hardware canary
+for the `v0.33.0` bump ([#326](https://github.com/88plug/k3d-gpu/pull/326)) found two
+causes on an Arc Pro B70 on 2026-11-07; the `v0.32.0` image exits there on the first one
+too:
+
+- The Service is named `vllm`, so Kubernetes gives the pod
+  `VLLM_PORT=tcp://<cluster IP>:8000`. vLLM reads `VLLM_PORT` as its own port and exits
+  at engine start ("appears to be a URI"), and the pod crash-loops.
+- The warm-up posts through the API server's service proxy with no `Content-Type`, and
+  vLLM answers such a request with 400.
+
+[#339](https://github.com/88plug/k3d-gpu/pull/339) fixed both in 0.1.94: the pod spec
+sets `enableServiceLinks: false`, and the warm-up runs curl inside the pod with
+`Content-Type: application/json`. llm-scaler and the Gaudi image are built on vLLM 0.26
+and 0.30, whose upstream source has both checks, and they get the same pod spec and
+warm-up; neither has run on hardware yet. Tensor parallelism above 1 has not run on
+Arc hardware either (`CCL_ZE_IPC_EXCHANGE=pidfd`; oneCCL calls
+`prctl(PR_SET_PTRACER, getppid())`, so sibling workers should need no `CAP_SYS_PTRACE`
+under yama `ptrace_scope=1`).
 
 | Variable | Default | Description |
 |----------|---------|-------------|
@@ -1039,7 +1069,11 @@ for tensor parallelism and Qwen models, `SYS_NICE`, an 8 GiB memory-backed
 bounds the HPU warm-up, which runs on every start and takes minutes;
 `K3D_GPU_AGENT_TIMEOUT` (1800 s) bounds it.
 
-Not yet run on Gaudi hardware here: serving end to end. The device plugin runs
+Not yet run on Gaudi hardware here: serving end to end. Releases up to 0.1.93 do not
+serve it: the Gaudi pod shares the Arc pod spec and warm-up, and with them the
+[two faults above](#vllm-backends-for-discrete-arc-battlemage) that
+[#339](https://github.com/88plug/k3d-gpu/pull/339) fixed in 0.1.94, whose end-to-end
+check ran on an Arc Pro B70 only. The device plugin runs
 without Intel's `habana-container-runtime` hook (see `share/gaudi-device-plugin.yml`),
 so whether vLLM opens the HPU with only the plugin's device nodes and environment
 is unverified. vllm-gaudi validates `Qwen3-0.6B`, not `Qwen3-4B`, and tensor
diff --git a/bench/vllm-agent/results/2026-11-08/README.md b/bench/vllm-agent/results/2026-11-08/README.md
new file mode 100644
index 00000000..4b8c62b5
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/README.md
@@ -0,0 +1,77 @@
+# The vLLM agent on an Arc Pro B70, unpatched (2026-11-08)
+
+The end state of `K3D_GPU_AGENT_BACKEND=vllm k3d-gpu agent up` on one Arc Pro B70, through
+the launcher alone: no `kubectl patch`, no warm-up sent by hand, the default image and
+model. Two launchers ran in one hold:
+
+- `main-0.1.94`: release 0.1.94, the first release with
+  [#339](https://github.com/88plug/k3d-gpu/pull/339) (no Service links in the pod, and
+  the warm-up sent from inside the pod as JSON).
+- `pr357-0985b179`: the head of [#357](https://github.com/88plug/k3d-gpu/pull/357), which
+  moves the image to `v0.33.0`. That head was regenerated onto main after #339 merged.
+
+Both passed. Releases 0.1.21 through 0.1.93 do not serve with vLLM: their Service gives
+the pod `VLLM_PORT=tcp://...`, on which vLLM exits, and their warm-up reaches vLLM with no
+`Content-Type`, which it answers with 400.
+
+This record covers vLLM XPU on Arc only. The Gaudi path and llm-scaler have not run on
+hardware, and neither has tensor parallelism above 1.
+
+## Host
+
+crescent: Xeon 6741P, 4x Arc Pro B70 on the `xe` driver, kernel `7.0.0-31-generic`, Ubuntu
+26.04.1 LTS. The hold took `~/k3dgpu-bench.lock`. It freed the cards by stopping
+`qwen38-a-hr`, `qwen38-b-hr` and `aiperf010` with `docker stop -t 120`, after a snapshot,
+a check that neither server had a request in the last 10 minutes, and each server's
+baseline answer ("Paris"). Afterwards it started them again, and both servers gave the same
+answer. All four cards read 0 GiB before each run and after each teardown.
+
+## Launchers and versions
+
+| | `main-0.1.94` | `pr357-0985b179` |
+|---|---|---|
+| Launcher | `/usr/bin/k3d-gpu` 0.1.94 from `k3d-gpu_0.1.94-1_all.deb` (sha256 `60a1dc5bbc5e6f2a439c34c9ec19dc2c916dcd35c5bec9d46752ee2ebcfd8dec`, the `aur-v0.1.94` release asset) | `scripts/k3d-gpu` 0.1.95 from a clean checkout of `0985b17902e507b13d8460970fbe2a15051ac5c9` (sha256 `e2d97e1960a04516d83ddc1c21c4142a406a2be6797ca1f102bd5feb1b495438`) |
+| How installed | `packaging/install.sh` from main at `e5f55b29` (sha256 `62c30da91025f5af25557e5e49da8c496591d98f3fb0f9a393c673b65edbbc55`), which pins 0.1.94; it also put k3d v5.9.0 and kubectl v1.37.2 in `~/.local/bin` | `scripts/checkout.sh` (`git fetch --depth 1` of the commit, no changes) |
+| Node image | `cryptoandcoffee/k3d-gpu:0.1.94` (`sha256:8630e14556e70111a097dcf22d15a7484bc94fa19a2d4b04b311ec60778d438e`) | `cryptoandcoffee/k3d-gpu:0.1.95` (`sha256:65499cbe202dca63c28029e03c4eb5139f5de31f5f369d915494f06e6f4dd3fa`) |
+| Agent image | `vllm/vllm-openai-xpu:v0.32.0@sha256:20509fabe50dda62c09c0786d1a6effa299d3dbdddcc49331b93e65144652a8c`, vLLM 0.32.0 | `vllm/vllm-openai-xpu:v0.33.0@sha256:be3a9aea234be64ffc2816ac561c597e0f0b1f44a61a7f4912780e4d77450086`, vLLM 0.33.0 |
+| Model | `Qwen/Qwen3-4B` at `1cfa9a7208912126459214e8b04321603b3df60c`, float16, `hermes` tool parser, `qwen3` reasoning parser | the same |
+
+The release package and the k3d and kubectl binaries the installer added were removed after
+the hold (`install-0.1.94/after-uninstall.txt`).
+
+## Results
+
+| | `main-0.1.94` | `pr357-0985b179` |
+|---|---|---|
+| `k3d-gpu up` | rc 0 in 48 s, allocatable `gpu.intel.com/xe: 4` | rc 0 in 49 s, the same |
+| `agent up` | rc 0 in 255 s | rc 0 in 258 s |
+| Pod | Ready, 0 restarts, limits `gpu.intel.com/xe: 1`, `enableServiceLinks: false`, no `VLLM_PORT` in its environment, one render node (`renderD129`) | the same, `renderD128` |
+| VRAM while serving | 27.50 GiB on 0000:48:00.0, 0 on the other three cards | 27.30 GiB on 0000:12:00.0, 0 on the other three |
+| KV cache (vLLM's log) | 137,472 tokens | 131,776 tokens |
+| The launcher's warm-up line | `warm-up reply (2s): I am a large language model developed by Alibaba Cloud.` | the same |
+| "What is 17 times 23?" | `391` (ends `\boxed{391}`), finish `stop`, 1,426 completion tokens (1,283 reasoning), 22.7 s | the same |
+| Tool call | `get_weather` with arguments `{"city": "Paris"}`, finish `tool_calls`, 1.6 s | the same |
+| Teardown | `agent down` and `down` rc 0, 0 GiB on all four cards, no k3d containers left | the same |
+| `run.sh` verdict | PASS | PASS |
+
+Both requests went through the endpoint `agent up` prints
+(`kubectl -n k3d-gpu-agent port-forward svc/vllm 8000:8000`). The chat request had no
+system prompt, thinking left on, temperature 0 and `max_tokens` 4096. The tool request
+offered one function, `get_weather(city)`, with "What is the weather in Paris right now?"
+and `tool_choice: auto`. `run.sh` passes a run only when the pod is Ready with one
+`gpu.intel.com/xe`, the launcher printed its warm-up reply, the chat answer contains 391,
+and the tool call is `get_weather` with `city` Paris.
+
+## Files
+
+| Path | What |
+|---|---|
+| `hold.log` | The hold, in order: the lock, the services stopped, each run's driver lines, the restart and the baseline answers |
+| `takeover-restore.txt` | The stop and restart logs of the three services, and the names stopped |
+| `queue` | The runs the hold took, as `<label>=<launcher>` |
+| `install-0.1.94/` | The installer's hash and output, `~/.local/bin` and the package before and after, and the removal |
+| `main-0.1.94/`, `pr357-0985b179/` | Per run: `driver.log` (its lines of `hold.log`), `up.txt`, `agent-up.txt` (and `.clean.txt`, colour codes stripped), `allocatable.json`, `pods.txt`, `deploy-vllm.yaml`, `describe-pods.txt`, `models.json`, `chat-req.json` and `chat.json`, `tool-req.json` and `tool.json`, `vllm.log`, `vllm-previous.log` (no previous container: 0 restarts), `events.txt`, `pf.log`, `tools.txt`, `agent-down.txt`, `down.txt`, `k3d-containers-after.txt` (empty) |
+| `scripts/` | The scripts as they ran on crescent: `hold.sh` runs `takeover-stop.sh`, then `run.sh` for each queued launcher, then `restore.sh`. `install-release.sh`, `checkout.sh` and `uninstall-release.sh` set up and remove the launchers. The one change since the run is a shellcheck directive comment in `uninstall-release.sh` |
+
+The takeover snapshot (`docker inspect` of the three services and their recent logs) stays
+on crescent: it holds those services' own configuration.
diff --git a/bench/vllm-agent/results/2026-11-08/hold.log b/bench/vllm-agent/results/2026-11-08/hold.log
new file mode 100644
index 00000000..34b23b28
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/hold.log
@@ -0,0 +1,80 @@
+2026-11-08T05:51:59Z waiting for ~/k3dgpu-bench.lock
+2026-11-08T05:51:59Z lock taken
+2026-11-08T05:51:59Z vram before: card1 15.67;card2 14.97;card3 15.67;card4 14.97;
+2026-11-08T05:53:41Z qwen38-a-hr POST lines in the last 10 min: 0
+2026-11-08T05:55:21Z qwen38-b-hr POST lines in the last 10 min: 0
+2026-11-08T05:55:21Z baseline port=30000 http=200 total_s=0.154155 {"model":"Qwen3.8-27B","content":"Paris","finish":"stop"}
+2026-11-08T05:55:21Z baseline port=30001 http=200 total_s=0.152700 {"model":"Qwen3.8-27B","content":"Paris","finish":"stop"}
+2026-11-08T05:55:21Z aiperf010 stop begin
+2026-11-08T05:57:21Z aiperf010 stop took 120s: status=exited exit=137 restart=no
+2026-11-08T05:57:21Z qwen38-b-hr stop begin
+2026-11-08T05:57:27Z qwen38-b-hr stop took 6s: status=exited exit=0 restart=no
+2026-11-08T05:57:27Z qwen38-a-hr stop begin
+2026-11-08T05:57:31Z qwen38-a-hr stop took 4s: status=exited exit=0 restart=no
+2026-11-08T05:57:42Z vram after: card1 0.00;card2 0.00;card3 0.00;card4 0.00;
+2026-11-08T05:57:42Z stopped: aiperf010 qwen38-b-hr qwen38-a-hr 
+2026-11-08T05:57:42Z run main-0.1.94 with /usr/bin/k3d-gpu
+2026-11-08T05:57:42Z [main-0.1.94] launcher /usr/bin/k3d-gpu: k3d-gpu 0.1.94
+2026-11-08T05:57:42Z [main-0.1.94] vram pre: 0000:12:00.0 card1 0.00;0000:48:00.0 card2 0.00;0000:7c:00.0 card3 0.00;0000:96:00.0 card4 0.00;
+2026-11-08T05:58:30Z [main-0.1.94] up rc=0 in 48 s
+2026-11-08T05:58:30Z [main-0.1.94] kubeconfig /home/devcloud/.local/state/k3d-gpu/kubeconfig-gpu
+2026-11-08T05:58:30Z [main-0.1.94] allocatable {"cpu":"96","ephemeral-storage":"3538173550480","gpu.intel.com/xe":"4","hugepages-1Gi":"0","hugepages-2Mi":"0","memory":"527767144Ki","pods":"110"}
+2026-11-08T06:02:45Z [main-0.1.94] agent up rc=0 in 255 s
+2026-11-08T06:02:45Z [main-0.1.94] launcher warm-up line: warm-up reply (2s): I am a large language model developed by Alibaba Cloud.
+2026-11-08T06:02:45Z [main-0.1.94] pod Ready=True limits={"gpu.intel.com/xe":"1"}
+2026-11-08T06:02:45Z [main-0.1.94] pod image vllm/vllm-openai-xpu:v0.32.0@sha256:20509fabe50dda62c09c0786d1a6effa299d3dbdddcc49331b93e65144652a8c docker.io/vllm/vllm-openai-xpu@sha256:20509fabe50dda62c09c0786d1a6effa299d3dbdddcc49331b93e65144652a8c restarts=0
+2026-11-08T06:02:45Z [main-0.1.94] pod enableServiceLinks=false
+2026-11-08T06:02:45Z [main-0.1.94] pod render nodes: card2 renderD129 
+2026-11-08T06:02:45Z [main-0.1.94] pod VLLM_PORT lines: 0
+2026-11-08T06:02:45Z [main-0.1.94] vram serving: 0000:12:00.0 card1 0.00;0000:48:00.0 card2 27.50;0000:7c:00.0 card3 0.00;0000:96:00.0 card4 0.00;
+2026-11-08T06:02:45Z [main-0.1.94] endpoint command: KUBECONFIG=/home/devcloud/.local/state/k3d-gpu/kubeconfig-gpu kubectl -n k3d-gpu-agent port-forward svc/vllm 8000:8000
+2026-11-08T06:02:47Z [main-0.1.94] models: ["Qwen3-4B"]
+2026-11-08T06:03:10Z [main-0.1.94] chat http=200 total_s=22.742111 finish=stop usage={"prompt_tokens":18,"total_tokens":1444,"completion_tokens":1426,"prompt_tokens_details":null,"completion_tokens_details":{"reasoning_tokens":1283}}
+2026-11-08T06:03:10Z [main-0.1.94] chat content:   To calculate $ 17 \times 23 $, we can use the distributive property of multiplication:  $$ 17 \times 23 = 17 \times (20 + 3) = (17 \times 20) + (17 \times 3) $$  Now compute each part:  - $ 17 \times 20 = 340 $ - $ 17 \times 3 = 51 $  Add the results:  $$ 340 + 51 = 391 $$  Thus, the final answer is:  $$ \boxed{391} $$ 
+2026-11-08T06:03:10Z [main-0.1.94] chat reasoning chars: 3593
+2026-11-08T06:03:12Z [main-0.1.94] tool http=200 total_s=1.625913 finish=tool_calls tool_calls=[{"id":"chatcmpl-tool-a2da8300fffa47a3","type":"function","function":{"name":"get_weather","arguments":"{\"city\": \"Paris\"}"}}]
+2026-11-08T06:03:12Z [main-0.1.94] tool arguments parse: {"city":"Paris"}
+2026-11-08T06:03:12Z [main-0.1.94] agent down rc=0
+2026-11-08T06:03:16Z [main-0.1.94] down rc=0
+2026-11-08T06:03:21Z [main-0.1.94] vram after: 0000:12:00.0 card1 0.00;0000:48:00.0 card2 0.00;0000:7c:00.0 card3 0.00;0000:96:00.0 card4 0.00;
+2026-11-08T06:03:21Z [main-0.1.94] k3d containers after: 0
+2026-11-08T06:03:21Z [main-0.1.94] RESULT PASS
+2026-11-08T06:03:21Z run main-0.1.94 rc=0
+2026-11-08T06:23:01Z run pr357-0985b179 with /home/devcloud/k3dgpu-bench/k3d-gpu-pr357-0985b179/scripts/k3d-gpu
+2026-11-08T06:23:01Z [pr357-0985b179] launcher /home/devcloud/k3dgpu-bench/k3d-gpu-pr357-0985b179/scripts/k3d-gpu: k3d-gpu 0.1.95
+2026-11-08T06:23:01Z [pr357-0985b179] vram pre: 0000:12:00.0 card1 0.00;0000:48:00.0 card2 0.00;0000:7c:00.0 card3 0.00;0000:96:00.0 card4 0.00;
+2026-11-08T06:23:50Z [pr357-0985b179] up rc=0 in 49 s
+2026-11-08T06:23:50Z [pr357-0985b179] kubeconfig /home/devcloud/.local/state/k3d-gpu/kubeconfig-gpu
+2026-11-08T06:23:50Z [pr357-0985b179] allocatable {"cpu":"96","ephemeral-storage":"3538173550480","gpu.intel.com/xe":"4","hugepages-1Gi":"0","hugepages-2Mi":"0","memory":"527767144Ki","pods":"110"}
+2026-11-08T06:28:08Z [pr357-0985b179] agent up rc=0 in 258 s
+2026-11-08T06:28:08Z [pr357-0985b179] launcher warm-up line: warm-up reply (2s): I am a large language model developed by Alibaba Cloud.
+2026-11-08T06:28:08Z [pr357-0985b179] pod Ready=True limits={"gpu.intel.com/xe":"1"}
+2026-11-08T06:28:08Z [pr357-0985b179] pod image vllm/vllm-openai-xpu:v0.33.0@sha256:be3a9aea234be64ffc2816ac561c597e0f0b1f44a61a7f4912780e4d77450086 docker.io/vllm/vllm-openai-xpu@sha256:be3a9aea234be64ffc2816ac561c597e0f0b1f44a61a7f4912780e4d77450086 restarts=0
+2026-11-08T06:28:08Z [pr357-0985b179] pod enableServiceLinks=false
+2026-11-08T06:28:08Z [pr357-0985b179] pod render nodes: card1 renderD128 
+2026-11-08T06:28:08Z [pr357-0985b179] pod VLLM_PORT lines: 0
+2026-11-08T06:28:08Z [pr357-0985b179] vram serving: 0000:12:00.0 card1 27.30;0000:48:00.0 card2 0.00;0000:7c:00.0 card3 0.00;0000:96:00.0 card4 0.00;
+2026-11-08T06:28:08Z [pr357-0985b179] endpoint command: KUBECONFIG=/home/devcloud/.local/state/k3d-gpu/kubeconfig-gpu kubectl -n k3d-gpu-agent port-forward svc/vllm 8000:8000
+2026-11-08T06:28:11Z [pr357-0985b179] models: ["Qwen3-4B"]
+2026-11-08T06:28:33Z [pr357-0985b179] chat http=200 total_s=22.721185 finish=stop usage={"prompt_tokens":18,"total_tokens":1444,"completion_tokens":1426,"prompt_tokens_details":null,"completion_tokens_details":{"reasoning_tokens":1283}}
+2026-11-08T06:28:33Z [pr357-0985b179] chat content:   To calculate $ 17 \times 23 $, we can use the distributive property of multiplication:  $$ 17 \times 23 = 17 \times (20 + 3) = (17 \times 20) + (17 \times 3) $$  Now compute each part:  - $ 17 \times 20 = 340 $ - $ 17 \times 3 = 51 $  Add the results:  $$ 340 + 51 = 391 $$  Thus, the final answer is:  $$ \boxed{391} $$ 
+2026-11-08T06:28:33Z [pr357-0985b179] chat reasoning chars: 3593
+2026-11-08T06:28:35Z [pr357-0985b179] tool http=200 total_s=1.623996 finish=tool_calls tool_calls=[{"id":"chatcmpl-tool-8874486735506104","type":"function","function":{"name":"get_weather","arguments":"{\"city\": \"Paris\"}"}}]
+2026-11-08T06:28:35Z [pr357-0985b179] tool arguments parse: {"city":"Paris"}
+2026-11-08T06:28:35Z [pr357-0985b179] agent down rc=0
+2026-11-08T06:28:39Z [pr357-0985b179] down rc=0
+2026-11-08T06:28:44Z [pr357-0985b179] vram after: 0000:12:00.0 card1 0.00;0000:48:00.0 card2 0.00;0000:7c:00.0 card3 0.00;0000:96:00.0 card4 0.00;
+2026-11-08T06:28:44Z [pr357-0985b179] k3d containers after: 0
+2026-11-08T06:28:44Z [pr357-0985b179] RESULT PASS
+2026-11-08T06:28:44Z run pr357-0985b179 rc=0
+2026-11-08T06:29:04Z release asked
+2026-11-08T06:29:04Z qwen38-a-hr start rc=0
+2026-11-08T06:29:05Z qwen38-b-hr start rc=0
+2026-11-08T06:29:05Z aiperf010 start rc=0
+2026-11-08T06:30:05Z OK qwen38-a-hr port=30000 http=200 total_s=0.386094 {"model":"Qwen3.8-27B","content":"Paris","finish":"stop","usage":{"prompt_tokens":24,"total_tokens":26,"completion_tokens":2,"prompt_tokens_details":null,"reasoning_tokens":0}} after 60s
+2026-11-08T06:30:06Z OK qwen38-b-hr port=30001 http=200 total_s=0.374647 {"model":"Qwen3.8-27B","content":"Paris","finish":"stop","usage":{"prompt_tokens":24,"total_tokens":26,"completion_tokens":2,"prompt_tokens_details":null,"reasoning_tokens":0}} after 0s
+2026-11-08T06:30:11Z aiperf010 status=running started=2026-11-08T06:29:05.053702422Z restart=no
+2026-11-08T06:30:16Z vram after restore: card1 26.46;card2 15.09;card3 26.46;card4 15.09;
+2026-11-08T06:30:16Z restore rc=0
+2026-11-08T06:30:16Z restore rc=0
+2026-11-08T06:30:16Z lock released
diff --git a/bench/vllm-agent/results/2026-11-08/install-0.1.94/after-uninstall.txt b/bench/vllm-agent/results/2026-11-08/install-0.1.94/after-uninstall.txt
new file mode 100644
index 00000000..b6875b50
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/install-0.1.94/after-uninstall.txt
@@ -0,0 +1,9 @@
+total 61836
+drwxrwxr-x 2 devcloud devcloud     4096 Nov  8 06:32 .
+drwxrwxr-x 6 devcloud devcloud     4096 Sep 23 07:50 ..
+-rw-rw-r-- 1 devcloud devcloud      328 Jul 31 00:11 env
+-rw-rw-r-- 1 devcloud devcloud      165 Jul 31 00:11 env.fish
+-rwxrwxr-x 1 devcloud devcloud  7842312 Sep 23 07:50 py-spy
+-rwxr-xr-x 1 devcloud devcloud 55110904 Jul 28 18:37 uv
+-rwxr-xr-x 1 devcloud devcloud   345952 Jul 28 18:37 uvx
+dpkg-query: no packages found matching k3d-gpu
diff --git a/bench/vllm-agent/results/2026-11-08/install-0.1.94/after.txt b/bench/vllm-agent/results/2026-11-08/install-0.1.94/after.txt
new file mode 100644
index 00000000..a5f0a389
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/install-0.1.94/after.txt
@@ -0,0 +1,16 @@
+total 146588
+drwxrwxr-x 2 devcloud devcloud     4096 Nov  8 05:51 .
+drwxrwxr-x 6 devcloud devcloud     4096 Sep 23 07:50 ..
+-rw-rw-r-- 1 devcloud devcloud      328 Jul 31 00:11 env
+-rw-rw-r-- 1 devcloud devcloud      165 Jul 31 00:11 env.fish
+-rwxr-xr-x 1 devcloud devcloud 24887458 Nov  8 05:51 k3d
+-rwxr-xr-x 1 devcloud devcloud 61890722 Nov  8 05:51 kubectl
+-rwxrwxr-x 1 devcloud devcloud  7842312 Sep 23 07:50 py-spy
+-rwxr-xr-x 1 devcloud devcloud 55110904 Jul 28 18:37 uv
+-rwxr-xr-x 1 devcloud devcloud   345952 Jul 28 18:37 uvx
++-- Desired=Unknown/Install/Remove/Purge/Hold
+|+- Status=Not/Inst/Conf-files/Unpacked/halF-conf/Half-inst/trig-aWait/Trig-pend
+||+ Err?=(none)/Reinst-required (Status,Err: uppercase=bad)
+||| Name           Version      Architecture Description
++++-==============-============-============-===========================================
+ii  k3d-gpu        0.1.94-1     all          Bootstrap GPU-ready k3d clusters on Docker.
diff --git a/bench/vllm-agent/results/2026-11-08/install-0.1.94/before.txt b/bench/vllm-agent/results/2026-11-08/install-0.1.94/before.txt
new file mode 100644
index 00000000..7f75366b
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/install-0.1.94/before.txt
@@ -0,0 +1,9 @@
+total 61836
+drwxrwxr-x 2 devcloud devcloud     4096 Sep 23 07:50 .
+drwxrwxr-x 6 devcloud devcloud     4096 Sep 23 07:50 ..
+-rw-rw-r-- 1 devcloud devcloud      328 Jul 31 00:11 env
+-rw-rw-r-- 1 devcloud devcloud      165 Jul 31 00:11 env.fish
+-rwxrwxr-x 1 devcloud devcloud  7842312 Sep 23 07:50 py-spy
+-rwxr-xr-x 1 devcloud devcloud 55110904 Jul 28 18:37 uv
+-rwxr-xr-x 1 devcloud devcloud   345952 Jul 28 18:37 uvx
+dpkg-query: no packages found matching k3d-gpu
diff --git a/bench/vllm-agent/results/2026-11-08/install-0.1.94/install.log b/bench/vllm-agent/results/2026-11-08/install-0.1.94/install.log
new file mode 100644
index 00000000..b27f5000
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/install-0.1.94/install.log
@@ -0,0 +1,63 @@
+==> k3d-gpu 0.1.94: k3d-gpu_0.1.94-1_all.deb
+==> sha256 ok: 60a1dc5bbc5e6f2a439c34c9ec19dc2c916dcd35c5bec9d46752ee2ebcfd8dec
+Reading package lists...
+Building dependency tree...
+Reading state information...
+Solving dependencies...
+The following packages were automatically installed and are no longer required:
+  libopenipmi0t64 libsnmp-base libsnmp40t64 libxpum1 openipmi pollinate
+Use 'sudo apt autoremove' to remove them.
+The following NEW packages will be installed:
+  k3d-gpu
+0 upgraded, 1 newly installed, 0 to remove and 83 not upgraded.
+Need to get 0 B/685 kB of archives.
+After this operation, 2688 kB of additional disk space will be used.
+Get:1 /tmp/tmp.5Pr8hESDWe/k3d-gpu_0.1.94-1_all.deb k3d-gpu all 0.1.94-1 [685 kB]
+debconf: unable to initialize frontend: Dialog
+debconf: (Dialog frontend will not work on a dumb terminal, an emacs shell buffer, or without a controlling terminal.)
+debconf: falling back to frontend: Readline
+debconf: unable to initialize frontend: Readline
+debconf: (This frontend requires a controlling tty.)
+debconf: falling back to frontend: Teletype
+debconf: unable to initialize frontend: Teletype
+debconf: (This frontend requires a controlling tty.)
+debconf: falling back to frontend: Noninteractive
+Selecting previously unselected package k3d-gpu.
+(Reading database ... 
(Reading database ... 5%
(Reading database ... 10%
(Reading database ... 15%
(Reading database ... 20%
(Reading database ... 25%
(Reading database ... 30%
(Reading database ... 35%
(Reading database ... 40%
(Reading database ... 45%
(Reading database ... 50%
(Reading database ... 55%
(Reading database ... 60%
(Reading database ... 65%
(Reading database ... 70%
(Reading database ... 75%
(Reading database ... 80%
(Reading database ... 85%
(Reading database ... 90%
(Reading database ... 95%
(Reading database ... 100%
(Reading database ... 220026 files and directories currently installed.)
+Preparing to unpack .../k3d-gpu_0.1.94-1_all.deb ...
+Unpacking k3d-gpu (0.1.94-1) ...
+Setting up k3d-gpu (0.1.94-1) ...
+debconf: unable to initialize frontend: Dialog
+debconf: (Dialog frontend will not work on a dumb terminal, an emacs shell buffer, or without a controlling terminal.)
+debconf: falling back to frontend: Readline
+debconf: unable to initialize frontend: Readline
+debconf: (This frontend requires a controlling tty.)
+debconf: falling back to frontend: Teletype
+debconf: unable to initialize frontend: Teletype
+debconf: (This frontend requires a controlling tty.)
+debconf: falling back to frontend: Noninteractive
+
+The processor microcode seems to be up-to-date.
+
+Restarting services...
+
+Service restarts being deferred:
+ /etc/needrestart/restart.d/dbus.service
+ systemctl restart networkd-dispatcher.service
+ systemctl restart systemd-logind.service
+ systemctl restart unattended-upgrades.service
+
+No containers need to be restarted.
+
+User sessions running outdated binaries:
+ devcloud @ session #1624: python3[965970]
+ devcloud @ session #4446: python3[1307983]
+ devcloud @ session #9941: tmux: server[2186548]
+ devcloud @ user manager: (sd-pam)[656124]
+ devcloud @ user service: dbus.service[2186553]
+
+No VM guests are running outdated hypervisor (qemu) binaries on this host.
+==> k3d: https://github.com/k3d-io/k3d/releases/download/v5.9.0/k3d-linux-amd64
+==> kubectl: https://dl.k8s.io/release/v1.37.2/bin/linux/amd64/kubectl
+==> add /home/devcloud/.local/bin to PATH: export PATH="/home/devcloud/.local/bin:$PATH"
+==> next: k3d-gpu doctor, then k3d-gpu quickstart
diff --git a/bench/vllm-agent/results/2026-11-08/install-0.1.94/install.sh.sha256 b/bench/vllm-agent/results/2026-11-08/install-0.1.94/install.sh.sha256
new file mode 100644
index 00000000..34ebb481
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/install-0.1.94/install.sh.sha256
@@ -0,0 +1 @@
+62c30da91025f5af25557e5e49da8c496591d98f3fb0f9a393c673b65edbbc55  install.sh
diff --git a/bench/vllm-agent/results/2026-11-08/install-0.1.94/uninstall.log b/bench/vllm-agent/results/2026-11-08/install-0.1.94/uninstall.log
new file mode 100644
index 00000000..543b1c46
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/install-0.1.94/uninstall.log
@@ -0,0 +1,13 @@
+Reading package lists...
+Building dependency tree...
+Reading state information...
+Solving dependencies...
+The following packages were automatically installed and are no longer required:
+  libopenipmi0t64 libsnmp-base libsnmp40t64 libxpum1 openipmi pollinate
+Use 'sudo apt autoremove' to remove them.
+The following packages will be REMOVED:
+  k3d-gpu
+0 upgraded, 0 newly installed, 1 to remove and 83 not upgraded.
+After this operation, 2688 kB disk space will be freed.
+(Reading database ... 
(Reading database ... 5%
(Reading database ... 10%
(Reading database ... 15%
(Reading database ... 20%
(Reading database ... 25%
(Reading database ... 30%
(Reading database ... 35%
(Reading database ... 40%
(Reading database ... 45%
(Reading database ... 50%
(Reading database ... 55%
(Reading database ... 60%
(Reading database ... 65%
(Reading database ... 70%
(Reading database ... 75%
(Reading database ... 80%
(Reading database ... 85%
(Reading database ... 90%
(Reading database ... 95%
(Reading database ... 100%
(Reading database ... 220126 files and directories currently installed.)
+Removing k3d-gpu (0.1.94-1) ...
diff --git a/bench/vllm-agent/results/2026-11-08/main-0.1.94/agent-down.txt b/bench/vllm-agent/results/2026-11-08/main-0.1.94/agent-down.txt
new file mode 100644
index 00000000..49cce86e
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/main-0.1.94/agent-down.txt
@@ -0,0 +1,4 @@
+deployment.apps "vllm" deleted from k3d-gpu-agent namespace
+service "vllm" deleted from k3d-gpu-agent namespace
+No resources found
+==> Model caches kept in the namespace's PVCs. Remove everything: KUBECONFIG=/home/devcloud/.local/state/k3d-gpu/kubeconfig-gpu kubectl delete namespace k3d-gpu-agent
diff --git a/bench/vllm-agent/results/2026-11-08/main-0.1.94/agent-up.clean.txt b/bench/vllm-agent/results/2026-11-08/main-0.1.94/agent-up.clean.txt
new file mode 100644
index 00000000..774cacd5
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/main-0.1.94/agent-up.clean.txt
@@ -0,0 +1,14 @@
+==> agent: vllm serving Qwen/Qwen3-4B as 'Qwen3-4B' on 1x gpu.intel.com/xe, 2 CPU(s)
+namespace/k3d-gpu-agent created
+persistentvolumeclaim/vllm-models created
+deployment.apps/vllm created
+service/vllm created
+==> Waiting for the model to download and load (first run fetches it into the PVC; up to 1800s)...
+Waiting for deployment "vllm" rollout to finish: 0 of 1 updated replicas are available...
+deployment "vllm" successfully rolled out
+==> warm-up reply (2s): I am a large language model developed by Alibaba Cloud.
+==> model Qwen/Qwen3-4B at revision 1cfa9a7208912126459214e8b04321603b3df60c
+==> OpenAI API ready, model 'Qwen3-4B':
+       KUBECONFIG=/home/devcloud/.local/state/k3d-gpu/kubeconfig-gpu kubectl -n k3d-gpu-agent port-forward svc/vllm 8000:8000
+       curl http://localhost:8000/v1/chat/completions -H 'Content-Type: application/json' \
+         -d '{"model":"Qwen3-4B","messages":[{"role":"user","content":"hi"}]}'
diff --git a/bench/vllm-agent/results/2026-11-08/main-0.1.94/agent-up.txt b/bench/vllm-agent/results/2026-11-08/main-0.1.94/agent-up.txt
new file mode 100644
index 00000000..d8bb73b7
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/main-0.1.94/agent-up.txt
@@ -0,0 +1,14 @@
+==> agent: vllm serving Qwen/Qwen3-4B as 'Qwen3-4B' on 1x gpu.intel.com/xe, 2 CPU(s)
+namespace/k3d-gpu-agent created
+persistentvolumeclaim/vllm-models created
+deployment.apps/vllm created
+service/vllm created
+==> Waiting for the model to download and load (first run fetches it into the PVC; up to 1800s)...
+Waiting for deployment "vllm" rollout to finish: 0 of 1 updated replicas are available...
+deployment "vllm" successfully rolled out
+==> warm-up reply (2s): I am a large language model developed by Alibaba Cloud.
+==> model Qwen/Qwen3-4B at revision 1cfa9a7208912126459214e8b04321603b3df60c
+==> OpenAI API ready, model 'Qwen3-4B':
+       KUBECONFIG=/home/devcloud/.local/state/k3d-gpu/kubeconfig-gpu kubectl -n k3d-gpu-agent port-forward svc/vllm 8000:8000
+       curl http://localhost:8000/v1/chat/completions -H 'Content-Type: application/json' \
+         -d '{"model":"Qwen3-4B","messages":[{"role":"user","content":"hi"}]}'
diff --git a/bench/vllm-agent/results/2026-11-08/main-0.1.94/allocatable.json b/bench/vllm-agent/results/2026-11-08/main-0.1.94/allocatable.json
new file mode 100644
index 00000000..db5960f2
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/main-0.1.94/allocatable.json
@@ -0,0 +1 @@
+{"cpu":"96","ephemeral-storage":"3538173550480","gpu.intel.com/xe":"4","hugepages-1Gi":"0","hugepages-2Mi":"0","memory":"527767144Ki","pods":"110"}
\ No newline at end of file
diff --git a/bench/vllm-agent/results/2026-11-08/main-0.1.94/chat-req.json b/bench/vllm-agent/results/2026-11-08/main-0.1.94/chat-req.json
new file mode 100644
index 00000000..d325a946
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/main-0.1.94/chat-req.json
@@ -0,0 +1,11 @@
+{
+  "model": "Qwen3-4B",
+  "messages": [
+    {
+      "role": "user",
+      "content": "What is 17 times 23?"
+    }
+  ],
+  "temperature": 0,
+  "max_tokens": 4096
+}
diff --git a/bench/vllm-agent/results/2026-11-08/main-0.1.94/chat.json b/bench/vllm-agent/results/2026-11-08/main-0.1.94/chat.json
new file mode 100644
index 00000000..e3fe7441
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/main-0.1.94/chat.json
@@ -0,0 +1 @@
+{"id":"chatcmpl-a3006ea2c1b336a6","object":"chat.completion","created":1794117767,"model":"Qwen3-4B","choices":[{"index":0,"message":{"role":"assistant","content":"\n\nTo calculate $ 17 \\times 23 $, we can use the distributive property of multiplication:\n\n$$\n17 \\times 23 = 17 \\times (20 + 3) = (17 \\times 20) + (17 \\times 3)\n$$\n\nNow compute each part:\n\n- $ 17 \\times 20 = 340 $\n- $ 17 \\times 3 = 51 $\n\nAdd the results:\n\n$$\n340 + 51 = 391\n$$\n\nThus, the final answer is:\n\n$$\n\\boxed{391}\n$$","refusal":null,"annotations":null,"audio":null,"function_call":null,"reasoning":"\nOkay, so I need to figure out what 17 multiplied by 23 is. Hmm, let me think. I remember that multiplication can be done in different ways, like breaking it down into smaller parts or using some tricks. Let me try the standard method first.\n\nFirst, I know that 17 times 20 is 340, right? Because 17 times 10 is 170, so times 2 would be 340. Then, 17 times 3 is 51. So if I add those two together, 340 plus 51, that should be 391. Wait, is that right? Let me check again.\n\nAlternatively, maybe I can use the distributive property. 17 times 23 can be thought of as (20 + 7) times (20 + 3). But that might be more complicated. Wait, maybe another way. Let me try multiplying 17 by 23 step by step.\n\nStarting with 17 multiplied by 23. Let me write it out vertically:\n\n   17\nx23\n-----\nFirst, multiply 17 by 3 (the units place of 23). 17 times 3 is 51. Then, multiply 17 by 2 (the tens place of 23), which is 34, but since it's in the tens place, it's actually 340. Then add them together: 51 + 340 = 391. So that's the same answer as before.\n\nWait, but maybe I should verify this another way to be sure. Let me try using another method. For example, 23 times 17. Maybe I can think of 23 as 20 + 3, so 20*17 + 3*17. 20*17 is 340, and 3*17 is 51. Adding them gives 340 + 51 = 391. Same result.\n\nAlternatively, maybe using the standard multiplication algorithm. Let me write it again:\n\n   17\nx23\n-----\n   51  (17 x 3)\n+340   (17 x 20)\n-----\n  391\n\nYes, that seems correct. But maybe I can check with another approach. Let's see, 17 times 23. If I think of 23 as 25 - 2, then 17 times 25 is 425, and then subtract 17 times 2, which is 34. So 425 - 34 = 391. That's another way to get the same answer.\n\nAlternatively, using the formula (a + b)(a - b) = a² - b², but that might not apply here. Wait, 17 and 23 are not symmetric around a central number. Let me see. Maybe not helpful here.\n\nAlternatively, maybe using the fact that 17 is 10 + 7, so 10*23 + 7*23. 10*23 is 230, and 7*23 is 161. Adding them together: 230 + 161 = 391. Yep, same answer again.\n\nHmm, so all these different methods are giving me 391. That seems pretty consistent. Maybe I can check with a calculator? Wait, I don't have a calculator here, but maybe I can do a quick check with another multiplication.\n\nLet me try multiplying 23 by 17. Let's do 23*10 = 230, 23*7 = 161. 230 + 161 = 391. Yep, same result. So I think that's correct.\n\nAlternatively, maybe I can use the area model. Imagine a rectangle divided into two parts: one part is 10 by 23, and the other is 7 by 23. So the total area would be 10*23 + 7*23 = 230 + 161 = 391. That's the same as before.\n\nWait, another way: 17*23. Let me think of 17 as 15 + 2. Then 15*23 + 2*23. 15*23 is 345, and 2*23 is 46. Adding them gives 345 + 46 = 391. Again, same answer.\n\nI think I'm confident that 17 times 23 is 391. All the different methods I tried lead to the same result, so I don't think I made a mistake here. Maybe I can check with a different approach, like using the fact that 17*20 is 340, and then 17*3 is 51, so 340 + 51 is 391. Yep, that's the same as before.\n\nAlternatively, maybe using the standard multiplication steps again. Let me write it out:\n\nMultiply 17 by 3: 17*3 = 51. Then, multiply 17 by 2 (which is in the tens place), so 17*2 = 34, but since it's the tens place, it's 340. Then add 51 and 340: 51 + 340 = 391. Yep, that's the same.\n\nI think I've verified it multiple times. So I think the answer is definitely 391. I don't see any errors in my calculations. All the different methods I tried give the same result, so I'm pretty sure that's correct.\n"},"logprobs":null,"finish_reason":"stop","stop_reason":null,"token_ids":null,"routed_experts":null}],"service_tier":null,"system_fingerprint":"vllm-0.32.0-b848f2c0","usage":{"prompt_tokens":18,"total_tokens":1444,"completion_tokens":1426,"prompt_tokens_details":null,"completion_tokens_details":{"reasoning_tokens":1283}},"prompt_logprobs":null,"prompt_token_ids":null,"prompt_text":null,"kv_transfer_params":null,"ec_transfer_params":null,"metrics":null}
\ No newline at end of file
diff --git a/bench/vllm-agent/results/2026-11-08/main-0.1.94/deploy-vllm.yaml b/bench/vllm-agent/results/2026-11-08/main-0.1.94/deploy-vllm.yaml
new file mode 100644
index 00000000..5e6a2fca
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/main-0.1.94/deploy-vllm.yaml
@@ -0,0 +1,156 @@
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  annotations:
+    deployment.kubernetes.io/revision: "1"
+    kubectl.kubernetes.io/last-applied-configuration: |
+      {"apiVersion":"apps/v1","kind":"Deployment","metadata":{"annotations":{},"labels":{"app":"vllm"},"name":"vllm","namespace":"k3d-gpu-agent"},"spec":{"replicas":1,"selector":{"matchLabels":{"app":"vllm"}},"strategy":{"type":"Recreate"},"template":{"metadata":{"annotations":{"k3d-gpu.io/agent-backend":"vllm","k3d-gpu.io/agent-device":"GPU"},"labels":{"app":"vllm"}},"spec":{"containers":[{"args":["Qwen/Qwen3-4B","--revision=1cfa9a7208912126459214e8b04321603b3df60c","--served-model-name=Qwen3-4B","--port=8000","--dtype=float16","--block-size=64","--max-model-len=8192","--max-num-batched-tokens=2048","--gpu-memory-utilization=0.9","--enforce-eager","--no-enable-prefix-caching","--tensor-parallel-size=1","--enable-auto-tool-choice","--tool-call-parser=hermes","--reasoning-parser=qwen3"],"env":[{"name":"VLLM_NO_USAGE_STATS","value":"1"},{"name":"DO_NOT_TRACK","value":"1"},{"name":"VLLM_WORKER_MULTIPROC_METHOD","value":"spawn"},{"name":"VLLM_ALLOW_LONG_MAX_MODEL_LEN","value":"1"},{"name":"VLLM_TARGET_DEVICE","value":"xpu"},{"name":"ZE_FLAT_DEVICE_HIERARCHY","value":"FLAT"},{"name":"ONEAPI_DEVICE_SELECTOR","value":"level_zero:*"},{"name":"ZE_ENABLE_PCI_ID_DEVICE_ORDER","value":"1"},{"name":"SYCL_CACHE_PERSISTENT","value":"1"},{"name":"ZES_ENABLE_SYSMAN","value":"1"},{"name":"HF_HUB_DISABLE_XET","value":"1"},{"name":"HF_HUB_ENABLE_HF_TRANSFER","value":"0"},{"name":"CCL_PROCESS_LAUNCHER","value":"none"},{"name":"CCL_WORKER_OFFLOAD","value":"0"},{"name":"CCL_WORKER_AFFINITY","value":""},{"name":"CCL_ZE_IPC_EXCHANGE","value":"pidfd"},{"name":"CCL_ALLREDUCE","value":"ring"},{"name":"HF_HOME","value":"/models/huggingface"},{"name":"XDG_CACHE_HOME","value":"/models/.cache"},{"name":"TRITON_CACHE_DIR","value":"/models/.triton"},{"name":"TRITON_HOME","value":"/models/.triton"}],"image":"vllm/vllm-openai-xpu:v0.32.0@sha256:20509fabe50dda62c09c0786d1a6effa299d3dbdddcc49331b93e65144652a8c","imagePullPolicy":"IfNotPresent","name":"vllm","ports":[{"containerPort":8000,"name":"http"}],"readinessProbe":{"httpGet":{"path":"/health","port":"http"},"periodSeconds":5},"resources":{"limits":{"gpu.intel.com/xe":1},"requests":{"cpu":"2","memory":"8Gi"}},"volumeMounts":[{"mountPath":"/models","name":"models"},{"mountPath":"/dev/shm","name":"shm"}]}],"enableServiceLinks":false,"securityContext":{"supplementalGroups":[990]},"volumes":[{"name":"models","persistentVolumeClaim":{"claimName":"vllm-models"}},{"emptyDir":{"medium":"Memory","sizeLimit":"2Gi"},"name":"shm"}]}}}}
+  creationTimestamp: "2026-11-08T05:58:31Z"
+  generation: 1
+  labels:
+    app: vllm
+  name: vllm
+  namespace: k3d-gpu-agent
+  resourceVersion: "969"
+  uid: e2e1c8ee-44ec-47a6-9ef0-a17644a86f1d
+spec:
+  progressDeadlineSeconds: 600
+  replicas: 1
+  revisionHistoryLimit: 10
+  selector:
+    matchLabels:
+      app: vllm
+  strategy:
+    type: Recreate
+  template:
+    metadata:
+      annotations:
+        k3d-gpu.io/agent-backend: vllm
+        k3d-gpu.io/agent-device: GPU
+      labels:
+        app: vllm
+    spec:
+      containers:
+      - args:
+        - Qwen/Qwen3-4B
+        - --revision=1cfa9a7208912126459214e8b04321603b3df60c
+        - --served-model-name=Qwen3-4B
+        - --port=8000
+        - --dtype=float16
+        - --block-size=64
+        - --max-model-len=8192
+        - --max-num-batched-tokens=2048
+        - --gpu-memory-utilization=0.9
+        - --enforce-eager
+        - --no-enable-prefix-caching
+        - --tensor-parallel-size=1
+        - --enable-auto-tool-choice
+        - --tool-call-parser=hermes
+        - --reasoning-parser=qwen3
+        env:
+        - name: VLLM_NO_USAGE_STATS
+          value: "1"
+        - name: DO_NOT_TRACK
+          value: "1"
+        - name: VLLM_WORKER_MULTIPROC_METHOD
+          value: spawn
+        - name: VLLM_ALLOW_LONG_MAX_MODEL_LEN
+          value: "1"
+        - name: VLLM_TARGET_DEVICE
+          value: xpu
+        - name: ZE_FLAT_DEVICE_HIERARCHY
+          value: FLAT
+        - name: ONEAPI_DEVICE_SELECTOR
+          value: level_zero:*
+        - name: ZE_ENABLE_PCI_ID_DEVICE_ORDER
+          value: "1"
+        - name: SYCL_CACHE_PERSISTENT
+          value: "1"
+        - name: ZES_ENABLE_SYSMAN
+          value: "1"
+        - name: HF_HUB_DISABLE_XET
+          value: "1"
+        - name: HF_HUB_ENABLE_HF_TRANSFER
+          value: "0"
+        - name: CCL_PROCESS_LAUNCHER
+          value: none
+        - name: CCL_WORKER_OFFLOAD
+          value: "0"
+        - name: CCL_WORKER_AFFINITY
+        - name: CCL_ZE_IPC_EXCHANGE
+          value: pidfd
+        - name: CCL_ALLREDUCE
+          value: ring
+        - name: HF_HOME
+          value: /models/huggingface
+        - name: XDG_CACHE_HOME
+          value: /models/.cache
+        - name: TRITON_CACHE_DIR
+          value: /models/.triton
+        - name: TRITON_HOME
+          value: /models/.triton
+        image: vllm/vllm-openai-xpu:v0.32.0@sha256:20509fabe50dda62c09c0786d1a6effa299d3dbdddcc49331b93e65144652a8c
+        imagePullPolicy: IfNotPresent
+        name: vllm
+        ports:
+        - containerPort: 8000
+          name: http
+          protocol: TCP
+        readinessProbe:
+          failureThreshold: 3
+          httpGet:
+            path: /health
+            port: http
+            scheme: HTTP
+          periodSeconds: 5
+          successThreshold: 1
+          timeoutSeconds: 1
+        resources:
+          limits:
+            gpu.intel.com/xe: "1"
+          requests:
+            cpu: "2"
+            memory: 8Gi
+        terminationMessagePath: /dev/termination-log
+        terminationMessagePolicy: File
+        volumeMounts:
+        - mountPath: /models
+          name: models
+        - mountPath: /dev/shm
+          name: shm
+      dnsPolicy: ClusterFirst
+      enableServiceLinks: false
+      restartPolicy: Always
+      schedulerName: default-scheduler
+      securityContext:
+        supplementalGroups:
+        - 990
+      terminationGracePeriodSeconds: 30
+      volumes:
+      - name: models
+        persistentVolumeClaim:
+          claimName: vllm-models
+      - emptyDir:
+          medium: Memory
+          sizeLimit: 2Gi
+        name: shm
+status:
+  availableReplicas: 1
+  conditions:
+  - lastTransitionTime: "2026-11-08T06:02:43Z"
+    lastUpdateTime: "2026-11-08T06:02:43Z"
+    message: Deployment has minimum availability.
+    reason: MinimumReplicasAvailable
+    status: "True"
+    type: Available
+  - lastTransitionTime: "2026-11-08T05:58:31Z"
+    lastUpdateTime: "2026-11-08T06:02:43Z"
+    message: ReplicaSet "vllm-6d68b9c67c" has successfully progressed.
+    reason: NewReplicaSetAvailable
+    status: "True"
+    type: Progressing
+  observedGeneration: 1
+  readyReplicas: 1
+  replicas: 1
+  terminatingReplicas: 0
+  updatedReplicas: 1
diff --git a/bench/vllm-agent/results/2026-11-08/main-0.1.94/describe-pods.txt b/bench/vllm-agent/results/2026-11-08/main-0.1.94/describe-pods.txt
new file mode 100644
index 00000000..bd6c22cc
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/main-0.1.94/describe-pods.txt
@@ -0,0 +1,110 @@
+Name:             vllm-6d68b9c67c-whhxr
+Namespace:        k3d-gpu-agent
+Priority:         0
+Service Account:  default
+Node:             k3d-gpu-server-0/172.19.0.2
+Start Time:       Sun, 08 Nov 2026 05:58:36 +0000
+Labels:           app=vllm
+                  pod-template-hash=6d68b9c67c
+Annotations:      k3d-gpu.io/agent-backend: vllm
+                  k3d-gpu.io/agent-device: GPU
+Status:           Running
+IP:               10.42.0.11
+IPs:
+  IP:           10.42.0.11
+Controlled By:  ReplicaSet/vllm-6d68b9c67c
+Containers:
+  vllm:
+    Container ID:  containerd://2d911db9881fc20ac33e5c206b1eb7597540c5f686b1e2239a21d6dbeffb78b6
+    Image:         vllm/vllm-openai-xpu:v0.32.0@sha256:20509fabe50dda62c09c0786d1a6effa299d3dbdddcc49331b93e65144652a8c
+    Image ID:      docker.io/vllm/vllm-openai-xpu@sha256:20509fabe50dda62c09c0786d1a6effa299d3dbdddcc49331b93e65144652a8c
+    Port:          8000/TCP (http)
+    Host Port:     0/TCP (http)
+    Args:
+      Qwen/Qwen3-4B
+      --revision=1cfa9a7208912126459214e8b04321603b3df60c
+      --served-model-name=Qwen3-4B
+      --port=8000
+      --dtype=float16
+      --block-size=64
+      --max-model-len=8192
+      --max-num-batched-tokens=2048
+      --gpu-memory-utilization=0.9
+      --enforce-eager
+      --no-enable-prefix-caching
+      --tensor-parallel-size=1
+      --enable-auto-tool-choice
+      --tool-call-parser=hermes
+      --reasoning-parser=qwen3
+    State:          Running
+      Started:      Sun, 08 Nov 2026 06:00:49 +0000
+    Ready:          True
+    Restart Count:  0
+    Limits:
+      gpu.intel.com/xe:  1
+    Requests:
+      cpu:               2
+      gpu.intel.com/xe:  1
+      memory:            8Gi
+    Readiness:           http-get http://:http/health delay=0s timeout=1s period=5s successThreshold=1 failureThreshold=3
+    Environment:
+      VLLM_NO_USAGE_STATS:            1
+      DO_NOT_TRACK:                   1
+      VLLM_WORKER_MULTIPROC_METHOD:   spawn
+      VLLM_ALLOW_LONG_MAX_MODEL_LEN:  1
+      VLLM_TARGET_DEVICE:             xpu
+      ZE_FLAT_DEVICE_HIERARCHY:       FLAT
+      ONEAPI_DEVICE_SELECTOR:         level_zero:*
+      ZE_ENABLE_PCI_ID_DEVICE_ORDER:  1
+      SYCL_CACHE_PERSISTENT:          1
+      ZES_ENABLE_SYSMAN:              1
+      HF_HUB_DISABLE_XET:             1
+      HF_HUB_ENABLE_HF_TRANSFER:      0
+      CCL_PROCESS_LAUNCHER:           none
+      CCL_WORKER_OFFLOAD:             0
+      CCL_WORKER_AFFINITY:            
+      CCL_ZE_IPC_EXCHANGE:            pidfd
+      CCL_ALLREDUCE:                  ring
+      HF_HOME:                        /models/huggingface
+      XDG_CACHE_HOME:                 /models/.cache
+      TRITON_CACHE_DIR:               /models/.triton
+      TRITON_HOME:                    /models/.triton
+    Mounts:
+      /dev/shm from shm (rw)
+      /models from models (rw)
+      /var/run/secrets/kubernetes.io/serviceaccount from kube-api-access-zf6cf (ro)
+Conditions:
+  Type                        Status
+  PodReadyToStartContainers   True 
+  Initialized                 True 
+  Ready                       True 
+  ContainersReady             True 
+  PodScheduled                True 
+Volumes:
+  models:
+    Type:       PersistentVolumeClaim (a reference to a PersistentVolumeClaim in the same namespace)
+    ClaimName:  vllm-models
+    ReadOnly:   false
+  shm:
+    Type:       EmptyDir (a temporary directory that shares a pod's lifetime)
+    Medium:     Memory
+    SizeLimit:  2Gi
+  kube-api-access-zf6cf:
+    Type:                    Projected (a volume that contains injected data from multiple sources)
+    TokenExpirationSeconds:  3607
+    ConfigMapName:           kube-root-ca.crt
+    Optional:                false
+    DownwardAPI:             true
+QoS Class:                   Burstable
+Node-Selectors:              <none>
+Tolerations:                 node.kubernetes.io/not-ready:NoExecute op=Exists for 300s
+                             node.kubernetes.io/unreachable:NoExecute op=Exists for 300s
+Events:
+  Type     Reason     Age                 From               Message
+  ----     ------     ----                ----               -------
+  Normal   Scheduled  4m8s                default-scheduler  Successfully assigned k3d-gpu-agent/vllm-6d68b9c67c-whhxr to k3d-gpu-server-0
+  Normal   Pulling    4m8s                kubelet            spec.containers{vllm}: Pulling image "vllm/vllm-openai-xpu:v0.32.0@sha256:20509fabe50dda62c09c0786d1a6effa299d3dbdddcc49331b93e65144652a8c"
+  Normal   Pulled     116s                kubelet            spec.containers{vllm}: Successfully pulled image "vllm/vllm-openai-xpu:v0.32.0@sha256:20509fabe50dda62c09c0786d1a6effa299d3dbdddcc49331b93e65144652a8c" in 2m12.498s (2m12.498s including waiting). Image size: 5181851385 bytes.
+  Normal   Created    116s                kubelet            spec.containers{vllm}: Container created
+  Normal   Started    116s                kubelet            spec.containers{vllm}: Container started
+  Warning  Unhealthy  7s (x25 over 116s)  kubelet            spec.containers{vllm}: Readiness probe failed: Get "http://10.42.0.11:8000/health": dial tcp 10.42.0.11:8000: connect: connection refused
diff --git a/bench/vllm-agent/results/2026-11-08/main-0.1.94/down.txt b/bench/vllm-agent/results/2026-11-08/main-0.1.94/down.txt
new file mode 100644
index 00000000..06ca520f
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/main-0.1.94/down.txt
@@ -0,0 +1,7 @@
+==> Deleting k3d cluster 'gpu'...
+INFO[0000] Deleting cluster 'gpu'                       
+INFO[0003] Deleting cluster network 'k3d-gpu'           
+INFO[0003] Deleting 1 attached volumes...               
+INFO[0003] Removing cluster details from default kubeconfig... 
+INFO[0003] Removing standalone kubeconfig file (if there is one)... 
+INFO[0003] Successfully deleted cluster gpu!            
diff --git a/bench/vllm-agent/results/2026-11-08/main-0.1.94/driver.log b/bench/vllm-agent/results/2026-11-08/main-0.1.94/driver.log
new file mode 100644
index 00000000..08860012
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/main-0.1.94/driver.log
@@ -0,0 +1,25 @@
+2026-11-08T05:57:42Z [main-0.1.94] launcher /usr/bin/k3d-gpu: k3d-gpu 0.1.94
+2026-11-08T05:57:42Z [main-0.1.94] vram pre: 0000:12:00.0 card1 0.00;0000:48:00.0 card2 0.00;0000:7c:00.0 card3 0.00;0000:96:00.0 card4 0.00;
+2026-11-08T05:58:30Z [main-0.1.94] up rc=0 in 48 s
+2026-11-08T05:58:30Z [main-0.1.94] kubeconfig /home/devcloud/.local/state/k3d-gpu/kubeconfig-gpu
+2026-11-08T05:58:30Z [main-0.1.94] allocatable {"cpu":"96","ephemeral-storage":"3538173550480","gpu.intel.com/xe":"4","hugepages-1Gi":"0","hugepages-2Mi":"0","memory":"527767144Ki","pods":"110"}
+2026-11-08T06:02:45Z [main-0.1.94] agent up rc=0 in 255 s
+2026-11-08T06:02:45Z [main-0.1.94] launcher warm-up line: warm-up reply (2s): I am a large language model developed by Alibaba Cloud.
+2026-11-08T06:02:45Z [main-0.1.94] pod Ready=True limits={"gpu.intel.com/xe":"1"}
+2026-11-08T06:02:45Z [main-0.1.94] pod image vllm/vllm-openai-xpu:v0.32.0@sha256:20509fabe50dda62c09c0786d1a6effa299d3dbdddcc49331b93e65144652a8c docker.io/vllm/vllm-openai-xpu@sha256:20509fabe50dda62c09c0786d1a6effa299d3dbdddcc49331b93e65144652a8c restarts=0
+2026-11-08T06:02:45Z [main-0.1.94] pod enableServiceLinks=false
+2026-11-08T06:02:45Z [main-0.1.94] pod render nodes: card2 renderD129 
+2026-11-08T06:02:45Z [main-0.1.94] pod VLLM_PORT lines: 0
+2026-11-08T06:02:45Z [main-0.1.94] vram serving: 0000:12:00.0 card1 0.00;0000:48:00.0 card2 27.50;0000:7c:00.0 card3 0.00;0000:96:00.0 card4 0.00;
+2026-11-08T06:02:45Z [main-0.1.94] endpoint command: KUBECONFIG=/home/devcloud/.local/state/k3d-gpu/kubeconfig-gpu kubectl -n k3d-gpu-agent port-forward svc/vllm 8000:8000
+2026-11-08T06:02:47Z [main-0.1.94] models: ["Qwen3-4B"]
+2026-11-08T06:03:10Z [main-0.1.94] chat http=200 total_s=22.742111 finish=stop usage={"prompt_tokens":18,"total_tokens":1444,"completion_tokens":1426,"prompt_tokens_details":null,"completion_tokens_details":{"reasoning_tokens":1283}}
+2026-11-08T06:03:10Z [main-0.1.94] chat content:   To calculate $ 17 \times 23 $, we can use the distributive property of multiplication:  $$ 17 \times 23 = 17 \times (20 + 3) = (17 \times 20) + (17 \times 3) $$  Now compute each part:  - $ 17 \times 20 = 340 $ - $ 17 \times 3 = 51 $  Add the results:  $$ 340 + 51 = 391 $$  Thus, the final answer is:  $$ \boxed{391} $$ 
+2026-11-08T06:03:10Z [main-0.1.94] chat reasoning chars: 3593
+2026-11-08T06:03:12Z [main-0.1.94] tool http=200 total_s=1.625913 finish=tool_calls tool_calls=[{"id":"chatcmpl-tool-a2da8300fffa47a3","type":"function","function":{"name":"get_weather","arguments":"{\"city\": \"Paris\"}"}}]
+2026-11-08T06:03:12Z [main-0.1.94] tool arguments parse: {"city":"Paris"}
+2026-11-08T06:03:12Z [main-0.1.94] agent down rc=0
+2026-11-08T06:03:16Z [main-0.1.94] down rc=0
+2026-11-08T06:03:21Z [main-0.1.94] vram after: 0000:12:00.0 card1 0.00;0000:48:00.0 card2 0.00;0000:7c:00.0 card3 0.00;0000:96:00.0 card4 0.00;
+2026-11-08T06:03:21Z [main-0.1.94] k3d containers after: 0
+2026-11-08T06:03:21Z [main-0.1.94] RESULT PASS
diff --git a/bench/vllm-agent/results/2026-11-08/main-0.1.94/events.txt b/bench/vllm-agent/results/2026-11-08/main-0.1.94/events.txt
new file mode 100644
index 00000000..b688aaff
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/main-0.1.94/events.txt
@@ -0,0 +1,147 @@
+NAMESPACE                LAST SEEN   TYPE      REASON                           OBJECT                                                           MESSAGE
+node-feature-discovery   5m7s        Normal    Scheduled                        pod/nfd-master-74b44b767d-ln6t5                                  Successfully assigned node-feature-discovery/nfd-master-74b44b767d-ln6t5 to k3d-gpu-server-0
+kube-system              5m7s        Normal    Scheduled                        pod/metrics-server-778c4649b4-nnkpg                              Successfully assigned kube-system/metrics-server-778c4649b4-nnkpg to k3d-gpu-server-0
+kube-system              5m7s        Normal    Scheduled                        pod/coredns-577d995dff-jpf5p                                     Successfully assigned kube-system/coredns-577d995dff-jpf5p to k3d-gpu-server-0
+kube-system              5m8s        Normal    Scheduled                        pod/helm-install-gateway-api-crd-t7pdz                           Successfully assigned kube-system/helm-install-gateway-api-crd-t7pdz to k3d-gpu-server-0
+kube-system              5m7s        Normal    Scheduled                        pod/intel-gpu-plugin-jk6tb                                       Successfully assigned kube-system/intel-gpu-plugin-jk6tb to k3d-gpu-server-0
+node-feature-discovery   5m7s        Normal    Scheduled                        pod/nfd-worker-z78lm                                             Successfully assigned node-feature-discovery/nfd-worker-z78lm to k3d-gpu-server-0
+node-feature-discovery   5m7s        Normal    Scheduled                        pod/nfd-gc-8d4595cbc-6ndqp                                       Successfully assigned node-feature-discovery/nfd-gc-8d4595cbc-6ndqp to k3d-gpu-server-0
+k3d-gpu-agent            4m35s       Normal    Scheduled                        pod/vllm-6d68b9c67c-whhxr                                        Successfully assigned k3d-gpu-agent/vllm-6d68b9c67c-whhxr to k3d-gpu-server-0
+default                  5m12s       Normal    Starting                         node/k3d-gpu-server-0                                            
+kube-system              5m7s        Normal    Scheduled                        pod/local-path-provisioner-6858d854cf-ks9lc                      Successfully assigned kube-system/local-path-provisioner-6858d854cf-ks9lc to k3d-gpu-server-0
+default                  5m14s       Normal    CertificateExpirationOK          node/k3d-gpu-server-0                                            Node and Certificate Authority certificates managed by k3s are OK
+default                  5m14s       Normal    NodeReady                        node/k3d-gpu-server-0                                            Node k3d-gpu-server-0 status is now: NodeReady
+default                  5m14s       Normal    NodeAllocatableEnforced          node/k3d-gpu-server-0                                            Updated Node Allocatable limit across pods
+default                  5m14s       Normal    Starting                         node/k3d-gpu-server-0                                            Starting kubelet.
+default                  5m14s       Normal    NodeHasSufficientPID             node/k3d-gpu-server-0                                            Node k3d-gpu-server-0 status is now: NodeHasSufficientPID
+default                  5m14s       Normal    NodeHasNoDiskPressure            node/k3d-gpu-server-0                                            Node k3d-gpu-server-0 status is now: NodeHasNoDiskPressure
+default                  5m14s       Normal    NodeHasSufficientMemory          node/k3d-gpu-server-0                                            Node k3d-gpu-server-0 status is now: NodeHasSufficientMemory
+default                  5m14s       Warning   InvalidDiskCapacity              node/k3d-gpu-server-0                                            invalid capacity 0 on image filesystem
+default                  5m12s       Normal    NodePasswordValidationComplete   node/k3d-gpu-server-0                                            Deferred node password secret validation complete
+kube-system              5m11s       Normal    ApplyingManifest                 addon/gaudi-device-plugin                                        Applying manifest at "/var/lib/rancher/k3s/server/manifests/gaudi-device-plugin.yaml"
+kube-system              5m11s       Normal    ApplyJob                         helmchart/gateway-api-crd                                        Applying HelmChart from https://%{KUBERNETES_API}%/static/charts/gateway-api-crd-1.6.103.tgz using Job kube-system/helm-install-gateway-api-crd
+kube-system              5m11s       Normal    ApplyingManifest                 addon/ccm                                                        Applying manifest at "/var/lib/rancher/k3s/server/manifests/ccm.yaml"
+kube-system              5m11s       Normal    AppliedManifest                  addon/gateway-api-crd                                            Applied manifest at "/var/lib/rancher/k3s/server/manifests/gateway-api-crd.yaml"
+kube-system              5m11s       Normal    ApplyingManifest                 addon/gateway-api-crd                                            Applying manifest at "/var/lib/rancher/k3s/server/manifests/gateway-api-crd.yaml"
+kube-system              5m11s       Normal    AppliedManifest                  addon/coredns                                                    Applied manifest at "/var/lib/rancher/k3s/server/manifests/coredns.yaml"
+kube-system              5m11s       Normal    ApplyingManifest                 addon/coredns                                                    Applying manifest at "/var/lib/rancher/k3s/server/manifests/coredns.yaml"
+kube-system              5m11s       Normal    ApplyingManifest                 addon/intel-dsa-plugin                                           Applying manifest at "/var/lib/rancher/k3s/server/manifests/intel-dsa-plugin.yaml"
+kube-system              5m11s       Normal    AppliedManifest                  addon/gaudi-device-plugin                                        Applied manifest at "/var/lib/rancher/k3s/server/manifests/gaudi-device-plugin.yaml"
+kube-system              5m11s       Normal    AppliedManifest                  addon/ccm                                                        Applied manifest at "/var/lib/rancher/k3s/server/manifests/ccm.yaml"
+kube-system              5m10s       Normal    ApplyingManifest                 addon/intel-gpu-wsl-plugin                                       Applying manifest at "/var/lib/rancher/k3s/server/manifests/intel-gpu-wsl-plugin.yaml"
+kube-system              5m10s       Normal    AppliedManifest                  addon/metrics-server-service                                     Applied manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/metrics-server-service.yaml"
+kube-system              5m10s       Normal    ApplyingManifest                 addon/auth-reader                                                Applying manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/auth-reader.yaml"
+kube-system              5m10s       Normal    AppliedManifest                  addon/auth-delegator                                             Applied manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/auth-delegator.yaml"
+kube-system              5m10s       Normal    ApplyingManifest                 addon/metrics-server-deployment                                  Applying manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/metrics-server-deployment.yaml"
+default                  5m10s       Normal    Synced                           node/k3d-gpu-server-0                                            Node synced successfully
+kube-system              5m10s       Normal    AppliedManifest                  addon/metrics-apiservice                                         Applied manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/metrics-apiservice.yaml"
+kube-system              5m10s       Normal    ApplyingManifest                 addon/auth-delegator                                             Applying manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/auth-delegator.yaml"
+kube-system              5m10s       Normal    ApplyingManifest                 addon/metrics-apiservice                                         Applying manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/metrics-apiservice.yaml"
+kube-system              5m10s       Normal    AppliedManifest                  addon/local-storage                                              Applied manifest at "/var/lib/rancher/k3s/server/manifests/local-storage.yaml"
+kube-system              5m10s       Normal    ApplyingManifest                 addon/local-storage                                              Applying manifest at "/var/lib/rancher/k3s/server/manifests/local-storage.yaml"
+kube-system              5m10s       Normal    AppliedManifest                  addon/aggregated-metrics-reader                                  Applied manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/aggregated-metrics-reader.yaml"
+kube-system              5m10s       Normal    ApplyingManifest                 addon/aggregated-metrics-reader                                  Applying manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/aggregated-metrics-reader.yaml"
+kube-system              5m10s       Normal    AppliedManifest                  addon/metrics-server-deployment                                  Applied manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/metrics-server-deployment.yaml"
+kube-system              5m10s       Normal    ApplyingManifest                 addon/metrics-server-service                                     Applying manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/metrics-server-service.yaml"
+kube-system              5m10s       Normal    AppliedManifest                  addon/intel-dsa-plugin                                           Applied manifest at "/var/lib/rancher/k3s/server/manifests/intel-dsa-plugin.yaml"
+kube-system              5m10s       Normal    DeletingManifest                 addon/intel-gpu-dra                                              Deleting manifest at "/var/lib/rancher/k3s/server/manifests/intel-gpu-dra.yaml"
+kube-system              5m10s       Normal    AppliedManifest                  addon/intel-qat-plugin                                           Applied manifest at "/var/lib/rancher/k3s/server/manifests/intel-qat-plugin.yaml"
+kube-system              5m10s       Normal    ApplyingManifest                 addon/intel-qat-plugin                                           Applying manifest at "/var/lib/rancher/k3s/server/manifests/intel-qat-plugin.yaml"
+kube-system              5m10s       Normal    AppliedManifest                  addon/auth-reader                                                Applied manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/auth-reader.yaml"
+kube-system              5m10s       Normal    AppliedManifest                  addon/intel-npu-plugin                                           Applied manifest at "/var/lib/rancher/k3s/server/manifests/intel-npu-plugin.yaml"
+kube-system              5m10s       Normal    ApplyingManifest                 addon/intel-npu-plugin                                           Applying manifest at "/var/lib/rancher/k3s/server/manifests/intel-npu-plugin.yaml"
+kube-system              5m10s       Normal    AppliedManifest                  addon/intel-iaa-plugin                                           Applied manifest at "/var/lib/rancher/k3s/server/manifests/intel-iaa-plugin.yaml"
+kube-system              5m10s       Normal    ApplyingManifest                 addon/intel-iaa-plugin                                           Applying manifest at "/var/lib/rancher/k3s/server/manifests/intel-iaa-plugin.yaml"
+kube-system              5m10s       Normal    AppliedManifest                  addon/intel-gpu-wsl-plugin                                       Applied manifest at "/var/lib/rancher/k3s/server/manifests/intel-gpu-wsl-plugin.yaml"
+kube-system              5m10s       Normal    AppliedManifest                  addon/intel-gpu-plugin                                           Applied manifest at "/var/lib/rancher/k3s/server/manifests/intel-gpu-plugin.yaml"
+kube-system              5m10s       Normal    AppliedManifest                  addon/resource-reader                                            Applied manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/resource-reader.yaml"
+kube-system              5m10s       Normal    ApplyingManifest                 addon/resource-reader                                            Applying manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/resource-reader.yaml"
+kube-system              5m10s       Normal    ApplyingManifest                 addon/intel-gpu-plugin                                           Applying manifest at "/var/lib/rancher/k3s/server/manifests/intel-gpu-plugin.yaml"
+kube-system              5m10s       Normal    ApplyingManifest                 addon/node-feature-discovery                                     Applying manifest at "/var/lib/rancher/k3s/server/manifests/node-feature-discovery.yaml"
+kube-system              5m9s        Normal    Suspended                        job/helm-install-gateway-api-crd                                 Job suspended
+default                  5m9s        Normal    RegisteredNode                   node/k3d-gpu-server-0                                            Node k3d-gpu-server-0 event: Registered Node k3d-gpu-server-0 in Controller
+node-feature-discovery   5m8s        Normal    SuccessfulCreate                 replicaset/nfd-master-74b44b767d                                 Created pod: nfd-master-74b44b767d-ln6t5
+kube-system              5m8s        Warning   FailedMount                      pod/helm-install-gateway-api-crd-t7pdz                           MountVolume.SetUp failed for volume "kube-api-access-8kfmz" : configmap "kube-root-ca.crt" not found
+kube-system              5m8s        Normal    AppliedManifest                  addon/node-feature-discovery                                     Applied manifest at "/var/lib/rancher/k3s/server/manifests/node-feature-discovery.yaml"
+node-feature-discovery   5m8s        Warning   FailedCreatePodSandBox           pod/nfd-worker-z78lm                                             Failed to create pod sandbox: rpc error: code = Unknown desc = failed to setup network for sandbox "4ab0236d05293aa757f87ce2b1f42a07f4b6c769e5a4d4c1fcb369cc18a4a8be": plugin type="flannel" failed (add): failed to load flannel 'subnet.env' file: open /run/flannel/subnet.env: no such file or directory. Check the flannel pod log for this node.
+kube-system              5m8s        Normal    ApplyingManifest                 addon/nvidia-device-plugin                                       Applying manifest at "/var/lib/rancher/k3s/server/manifests/nvidia-device-plugin.yaml"
+kube-system              5m8s        Normal    AppliedManifest                  addon/nvidia-device-plugin                                       Applied manifest at "/var/lib/rancher/k3s/server/manifests/nvidia-device-plugin.yaml"
+kube-system              5m8s        Normal    DeletingManifest                 addon/traefik                                                    Deleting manifest at "/var/lib/rancher/k3s/server/manifests/traefik.yaml"
+kube-system              5m8s        Warning   FailedCreatePodSandBox           pod/intel-gpu-plugin-jk6tb                                       Failed to create pod sandbox: rpc error: code = Unknown desc = failed to setup network for sandbox "22f06b79c3fce0b8d0704369243abda3a2b8ada439dfd3cfdbea1aa3f387b794": plugin type="flannel" failed (add): failed to load flannel 'subnet.env' file: open /run/flannel/subnet.env: no such file or directory. Check the flannel pod log for this node.
+kube-system              5m8s        Normal    AppliedManifest                  addon/runtimes                                                   Applied manifest at "/var/lib/rancher/k3s/server/manifests/runtimes.yaml"
+kube-system              5m8s        Normal    Resumed                          job/helm-install-gateway-api-crd                                 Job resumed
+node-feature-discovery   5m8s        Normal    SuccessfulCreate                 daemonset/nfd-worker                                             Created pod: nfd-worker-z78lm
+kube-system              5m8s        Normal    ScalingReplicaSet                deployment/metrics-server                                        Scaled up replica set metrics-server-778c4649b4 from 0 to 1
+kube-system              5m8s        Normal    SuccessfulCreate                 job/helm-install-gateway-api-crd                                 Created pod: helm-install-gateway-api-crd-t7pdz
+kube-system              5m8s        Normal    ApplyingManifest                 addon/rolebindings                                               Applying manifest at "/var/lib/rancher/k3s/server/manifests/rolebindings.yaml"
+kube-system              5m8s        Normal    SuccessfulCreate                 daemonset/intel-gpu-plugin                                       Created pod: intel-gpu-plugin-jk6tb
+kube-system              5m8s        Normal    SuccessfulCreate                 replicaset/metrics-server-778c4649b4                             Created pod: metrics-server-778c4649b4-nnkpg
+kube-system              5m8s        Normal    AppliedManifest                  addon/rolebindings                                               Applied manifest at "/var/lib/rancher/k3s/server/manifests/rolebindings.yaml"
+kube-system              5m8s        Normal    ApplyingManifest                 addon/runtimes                                                   Applying manifest at "/var/lib/rancher/k3s/server/manifests/runtimes.yaml"
+kube-system              5m8s        Normal    SuccessfulCreate                 replicaset/coredns-577d995dff                                    Created pod: coredns-577d995dff-jpf5p
+kube-system              5m8s        Normal    DeletingManifest                 addon/nri-balloons                                               Deleting manifest at "/var/lib/rancher/k3s/server/manifests/nri-balloons.yaml"
+node-feature-discovery   5m8s        Normal    ScalingReplicaSet                deployment/nfd-master                                            Scaled up replica set nfd-master-74b44b767d from 0 to 1
+kube-system              5m8s        Normal    ScalingReplicaSet                deployment/local-path-provisioner                                Scaled up replica set local-path-provisioner-6858d854cf from 0 to 1
+kube-system              5m8s        Normal    ResumeJob                        helmchart/gateway-api-crd                                        Resumed synced Job kube-system/helm-install-gateway-api-crd
+kube-system              5m8s        Normal    SuccessfulCreate                 replicaset/local-path-provisioner-6858d854cf                     Created pod: local-path-provisioner-6858d854cf-ks9lc
+kube-system              5m8s        Normal    ScalingReplicaSet                deployment/coredns                                               Scaled up replica set coredns-577d995dff from 0 to 1
+node-feature-discovery   5m8s        Normal    SuccessfulCreate                 replicaset/nfd-gc-8d4595cbc                                      Created pod: nfd-gc-8d4595cbc-6ndqp
+node-feature-discovery   5m8s        Normal    ScalingReplicaSet                deployment/nfd-gc                                                Scaled up replica set nfd-gc-8d4595cbc from 0 to 1
+kube-system              5m7s        Warning   FailedCreatePodSandBox           pod/local-path-provisioner-6858d854cf-ks9lc                      Failed to create pod sandbox: rpc error: code = Unknown desc = failed to setup network for sandbox "2f8813ea9ba2ba196e827554d30feab958ce81f2634e8bd7c8f2cc661698a94b": plugin type="flannel" failed (add): failed to load flannel 'subnet.env' file: open /run/flannel/subnet.env: no such file or directory. Check the flannel pod log for this node.
+node-feature-discovery   5m7s        Warning   FailedCreatePodSandBox           pod/nfd-master-74b44b767d-ln6t5                                  Failed to create pod sandbox: rpc error: code = Unknown desc = failed to setup network for sandbox "30146729a5125725f0b8dab31bfac9603b7cb2df1f1ba24e32d2ba7f9c576d3a": plugin type="flannel" failed (add): failed to load flannel 'subnet.env' file: open /run/flannel/subnet.env: no such file or directory. Check the flannel pod log for this node.
+kube-system              5m7s        Warning   FailedCreatePodSandBox           pod/helm-install-gateway-api-crd-t7pdz                           Failed to create pod sandbox: rpc error: code = Unknown desc = failed to setup network for sandbox "706a34689a8133720ca7feda2fca9da9ac9c716659f178fe96ba738ef2baafb5": plugin type="flannel" failed (add): failed to load flannel 'subnet.env' file: open /run/flannel/subnet.env: no such file or directory. Check the flannel pod log for this node.
+node-feature-discovery   5m7s        Warning   FailedCreatePodSandBox           pod/nfd-gc-8d4595cbc-6ndqp                                       Failed to create pod sandbox: rpc error: code = Unknown desc = failed to setup network for sandbox "8dd665c4372ffdbe4a997917044ae3947475e62d3b92a79ea892e2209dc43dcc": plugin type="flannel" failed (add): failed to load flannel 'subnet.env' file: open /run/flannel/subnet.env: no such file or directory. Check the flannel pod log for this node.
+kube-system              5m7s        Warning   FailedCreatePodSandBox           pod/metrics-server-778c4649b4-nnkpg                              Failed to create pod sandbox: rpc error: code = Unknown desc = failed to setup network for sandbox "618229a9b4c295093ffde18ce39b237a5e2a74bc4882e5dcc4f1d14273e92650": plugin type="flannel" failed (add): failed to load flannel 'subnet.env' file: open /run/flannel/subnet.env: no such file or directory. Check the flannel pod log for this node.
+kube-system              5m7s        Warning   FailedCreatePodSandBox           pod/coredns-577d995dff-jpf5p                                     Failed to create pod sandbox: rpc error: code = Unknown desc = failed to setup network for sandbox "9e4fca51c744730aa581d2a31e47df285c12241e93f2c55fc0d57a08bcb8a79b": plugin type="flannel" failed (add): failed to load flannel 'subnet.env' file: open /run/flannel/subnet.env: no such file or directory. Check the flannel pod log for this node.
+kube-system              4m55s       Normal    Pulling                          pod/local-path-provisioner-6858d854cf-ks9lc                      Pulling image "rancher/local-path-provisioner:v0.0.37"
+node-feature-discovery   4m55s       Normal    Pulling                          pod/nfd-master-74b44b767d-ln6t5                                  Pulling image "registry.k8s.io/nfd/node-feature-discovery:v0.20.1@sha256:6fc5c9af9176e1acb5ce57a4a1d0c0d3eb25cd34f5bdd0efefebf0efe45a8379"
+kube-system              4m55s       Normal    Pulling                          pod/intel-gpu-plugin-jk6tb                                       Pulling image "intel/intel-gpu-plugin:0.37.1@sha256:004641b2547d58f1eadce58473fa160ad71377de1eb588cf1f82832387bf3e33"
+kube-system              4m55s       Normal    Pulling                          pod/metrics-server-778c4649b4-nnkpg                              Pulling image "rancher/mirrored-metrics-server:v0.9.0"
+node-feature-discovery   4m53s       Normal    Pulling                          pod/nfd-worker-z78lm                                             Pulling image "registry.k8s.io/nfd/node-feature-discovery:v0.20.1@sha256:6fc5c9af9176e1acb5ce57a4a1d0c0d3eb25cd34f5bdd0efefebf0efe45a8379"
+node-feature-discovery   4m53s       Normal    Pulling                          pod/nfd-gc-8d4595cbc-6ndqp                                       Pulling image "registry.k8s.io/nfd/node-feature-discovery:v0.20.1@sha256:6fc5c9af9176e1acb5ce57a4a1d0c0d3eb25cd34f5bdd0efefebf0efe45a8379"
+kube-system              4m52s       Normal    Created                          pod/intel-gpu-plugin-jk6tb                                       Container created
+kube-system              4m52s       Normal    Created                          pod/local-path-provisioner-6858d854cf-ks9lc                      Container created
+kube-system              4m52s       Normal    Pulling                          pod/coredns-577d995dff-jpf5p                                     Pulling image "rancher/mirrored-coredns-coredns:1.14.7"
+kube-system              4m52s       Normal    Pulling                          pod/helm-install-gateway-api-crd-t7pdz                           Pulling image "rancher/klipper-helm:v0.13.3-build20260727"
+kube-system              4m52s       Normal    Pulled                           pod/intel-gpu-plugin-jk6tb                                       Successfully pulled image "intel/intel-gpu-plugin:0.37.1@sha256:004641b2547d58f1eadce58473fa160ad71377de1eb588cf1f82832387bf3e33" in 2.89s (2.89s including waiting). Image size: 5970136 bytes.
+kube-system              4m52s       Normal    Pulled                           pod/local-path-provisioner-6858d854cf-ks9lc                      Successfully pulled image "rancher/local-path-provisioner:v0.0.37" in 2.305s (2.305s including waiting). Image size: 15066012 bytes.
+kube-system              4m52s       Normal    Started                          pod/local-path-provisioner-6858d854cf-ks9lc                      Container started
+node-feature-discovery   4m51s       Normal    Created                          pod/nfd-worker-z78lm                                             Container created
+node-feature-discovery   4m51s       Normal    Started                          pod/nfd-master-74b44b767d-ln6t5                                  Container started
+kube-system              4m51s       Normal    Started                          pod/intel-gpu-plugin-jk6tb                                       Container started
+kube-system              4m51s       Normal    Created                          pod/metrics-server-778c4649b4-nnkpg                              Container created
+kube-system              4m51s       Normal    Started                          pod/metrics-server-778c4649b4-nnkpg                              Container started
+node-feature-discovery   4m51s       Normal    Pulled                           pod/nfd-worker-z78lm                                             Successfully pulled image "registry.k8s.io/nfd/node-feature-discovery:v0.20.1@sha256:6fc5c9af9176e1acb5ce57a4a1d0c0d3eb25cd34f5bdd0efefebf0efe45a8379" in 2.045s (2.045s including waiting). Image size: 87537491 bytes.
+node-feature-discovery   4m51s       Warning   Unhealthy                        pod/nfd-master-74b44b767d-ln6t5                                  Readiness probe failed: Get "http://10.42.0.2:8080/healthz": dial tcp 10.42.0.2:8080: connect: connection refused
+kube-system              4m51s       Normal    Pulled                           pod/metrics-server-778c4649b4-nnkpg                              Successfully pulled image "rancher/mirrored-metrics-server:v0.9.0" in 3.2s (3.2s including waiting). Image size: 23258428 bytes.
+node-feature-discovery   4m51s       Normal    Created                          pod/nfd-master-74b44b767d-ln6t5                                  Container created
+node-feature-discovery   4m51s       Normal    Pulled                           pod/nfd-master-74b44b767d-ln6t5                                  Successfully pulled image "registry.k8s.io/nfd/node-feature-discovery:v0.20.1@sha256:6fc5c9af9176e1acb5ce57a4a1d0c0d3eb25cd34f5bdd0efefebf0efe45a8379" in 3.532s (3.532s including waiting). Image size: 87537491 bytes.
+node-feature-discovery   4m51s       Normal    Started                          pod/nfd-gc-8d4595cbc-6ndqp                                       Container started
+node-feature-discovery   4m51s       Normal    Created                          pod/nfd-gc-8d4595cbc-6ndqp                                       Container created
+node-feature-discovery   4m51s       Normal    Started                          pod/nfd-worker-z78lm                                             Container started
+node-feature-discovery   4m51s       Normal    Pulled                           pod/nfd-gc-8d4595cbc-6ndqp                                       Successfully pulled image "registry.k8s.io/nfd/node-feature-discovery:v0.20.1@sha256:6fc5c9af9176e1acb5ce57a4a1d0c0d3eb25cd34f5bdd0efefebf0efe45a8379" in 2.042s (2.042s including waiting). Image size: 87537491 bytes.
+kube-system              4m51s       Warning   Unhealthy                        pod/metrics-server-778c4649b4-nnkpg                              Readiness probe failed: Get "https://10.42.0.3:10250/readyz": dial tcp 10.42.0.3:10250: connect: connection refused
+kube-system              4m49s       Normal    Pulled                           pod/coredns-577d995dff-jpf5p                                     Successfully pulled image "rancher/mirrored-coredns-coredns:1.14.7" in 3.107s (3.107s including waiting). Image size: 24261571 bytes.
+kube-system              4m49s       Normal    Started                          pod/coredns-577d995dff-jpf5p                                     Container started
+kube-system              4m49s       Warning   Unhealthy                        pod/coredns-577d995dff-jpf5p                                     Readiness probe failed: Get "http://10.42.0.8:8181/ready": dial tcp 10.42.0.8:8181: connect: connection refused
+kube-system              4m49s       Normal    Pulled                           pod/helm-install-gateway-api-crd-t7pdz                           Successfully pulled image "rancher/klipper-helm:v0.13.3-build20260727" in 3.68s (3.68s including waiting). Image size: 74172006 bytes.
+kube-system              4m49s       Normal    Created                          pod/helm-install-gateway-api-crd-t7pdz                           Container created
+kube-system              4m49s       Normal    Created                          pod/coredns-577d995dff-jpf5p                                     Container created
+kube-system              4m48s       Normal    Started                          pod/helm-install-gateway-api-crd-t7pdz                           Container started
+kube-system              4m45s       Normal    Completed                        job/helm-install-gateway-api-crd                                 Job completed
+k3d-gpu-agent            4m41s       Normal    ScalingReplicaSet                deployment/vllm                                                  Scaled up replica set vllm-6d68b9c67c from 0 to 1
+k3d-gpu-agent            4m41s       Normal    WaitForFirstConsumer             persistentvolumeclaim/vllm-models                                waiting for first consumer to be created before binding
+k3d-gpu-agent            4m41s       Normal    Provisioning                     persistentvolumeclaim/vllm-models                                External provisioner is provisioning volume for claim "k3d-gpu-agent/vllm-models"
+k3d-gpu-agent            4m41s       Normal    SuccessfulCreate                 replicaset/vllm-6d68b9c67c                                       Created pod: vllm-6d68b9c67c-whhxr
+kube-system              4m40s       Normal    Pulling                          pod/helper-pod-create-pvc-e32d7467-d303-4dcc-835d-46fc9181e53b   Pulling image "rancher/mirrored-library-busybox:1.37.0"
+kube-system              4m39s       Normal    Created                          pod/helper-pod-create-pvc-e32d7467-d303-4dcc-835d-46fc9181e53b   Container created
+k3d-gpu-agent            4m39s       Normal    ExternalProvisioning             persistentvolumeclaim/vllm-models                                Waiting for a volume to be created either by the external provisioner 'rancher.io/local-path' or manually by the system administrator. If volume creation is delayed, please verify that the provisioner is running and correctly registered.
+kube-system              4m39s       Normal    Pulled                           pod/helper-pod-create-pvc-e32d7467-d303-4dcc-835d-46fc9181e53b   Successfully pulled image "rancher/mirrored-library-busybox:1.37.0" in 1.505s (1.505s including waiting). Image size: 2151353 bytes.
+kube-system              4m39s       Normal    Started                          pod/helper-pod-create-pvc-e32d7467-d303-4dcc-835d-46fc9181e53b   Container started
+k3d-gpu-agent            4m37s       Normal    ProvisioningSucceeded            persistentvolumeclaim/vllm-models                                Successfully provisioned volume pvc-e32d7467-d303-4dcc-835d-46fc9181e53b
+kube-system              4m37s       Warning   Unhealthy                        pod/metrics-server-778c4649b4-nnkpg                              Readiness probe failed: HTTP probe failed with statuscode: 500
+k3d-gpu-agent            4m35s       Normal    Pulling                          pod/vllm-6d68b9c67c-whhxr                                        Pulling image "vllm/vllm-openai-xpu:v0.32.0@sha256:20509fabe50dda62c09c0786d1a6effa299d3dbdddcc49331b93e65144652a8c"
+k3d-gpu-agent            2m23s       Normal    Created                          pod/vllm-6d68b9c67c-whhxr                                        Container created
+k3d-gpu-agent            2m23s       Normal    Pulled                           pod/vllm-6d68b9c67c-whhxr                                        Successfully pulled image "vllm/vllm-openai-xpu:v0.32.0@sha256:20509fabe50dda62c09c0786d1a6effa299d3dbdddcc49331b93e65144652a8c" in 2m12.498s (2m12.498s including waiting). Image size: 5181851385 bytes.
+k3d-gpu-agent            2m23s       Normal    Started                          pod/vllm-6d68b9c67c-whhxr                                        Container started
+k3d-gpu-agent            34s         Warning   Unhealthy                        pod/vllm-6d68b9c67c-whhxr                                        Readiness probe failed: Get "http://10.42.0.11:8000/health": dial tcp 10.42.0.11:8000: connect: connection refused
diff --git a/bench/vllm-agent/results/2026-11-08/main-0.1.94/k3d-containers-after.txt b/bench/vllm-agent/results/2026-11-08/main-0.1.94/k3d-containers-after.txt
new file mode 100644
index 00000000..e69de29b
diff --git a/bench/vllm-agent/results/2026-11-08/main-0.1.94/models.json b/bench/vllm-agent/results/2026-11-08/main-0.1.94/models.json
new file mode 100644
index 00000000..8dbb8b06
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/main-0.1.94/models.json
@@ -0,0 +1 @@
+{"object":"list","data":[{"id":"Qwen3-4B","object":"model","created":1794117767,"owned_by":"vllm","root":"Qwen/Qwen3-4B","parent":null,"max_model_len":8192,"permission":[{"id":"modelperm-a46e90aa762d2544","object":"model_permission","created":1794117767,"allow_create_engine":false,"allow_sampling":true,"allow_logprobs":true,"allow_search_indices":false,"allow_view":true,"allow_fine_tuning":false,"organization":"*","group":null,"is_blocking":false}]}]}
\ No newline at end of file
diff --git a/bench/vllm-agent/results/2026-11-08/main-0.1.94/pf.log b/bench/vllm-agent/results/2026-11-08/main-0.1.94/pf.log
new file mode 100644
index 00000000..44076ae3
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/main-0.1.94/pf.log
@@ -0,0 +1,5 @@
+Forwarding from 127.0.0.1:8000 -> 8000
+Forwarding from [::1]:8000 -> 8000
+Handling connection for 8000
+Handling connection for 8000
+Handling connection for 8000
diff --git a/bench/vllm-agent/results/2026-11-08/main-0.1.94/pods.txt b/bench/vllm-agent/results/2026-11-08/main-0.1.94/pods.txt
new file mode 100644
index 00000000..f7127a3d
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/main-0.1.94/pods.txt
@@ -0,0 +1,8 @@
+NAME                   READY   UP-TO-DATE   AVAILABLE   AGE     CONTAINERS   IMAGES                                                                                                 SELECTOR
+deployment.apps/vllm   1/1     1            1           4m14s   vllm         vllm/vllm-openai-xpu:v0.32.0@sha256:20509fabe50dda62c09c0786d1a6effa299d3dbdddcc49331b93e65144652a8c   app=vllm
+
+NAME                        READY   STATUS    RESTARTS   AGE     IP           NODE               NOMINATED NODE   READINESS GATES
+pod/vllm-6d68b9c67c-whhxr   1/1     Running   0          4m14s   10.42.0.11   k3d-gpu-server-0   <none>           <none>
+
+NAME           TYPE        CLUSTER-IP      EXTERNAL-IP   PORT(S)    AGE     SELECTOR
+service/vllm   ClusterIP   10.43.197.185   <none>        8000/TCP   4m14s   app=vllm
diff --git a/bench/vllm-agent/results/2026-11-08/main-0.1.94/tool-req.json b/bench/vllm-agent/results/2026-11-08/main-0.1.94/tool-req.json
new file mode 100644
index 00000000..af9846a3
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/main-0.1.94/tool-req.json
@@ -0,0 +1,33 @@
+{
+  "model": "Qwen3-4B",
+  "messages": [
+    {
+      "role": "user",
+      "content": "What is the weather in Paris right now?"
+    }
+  ],
+  "tools": [
+    {
+      "type": "function",
+      "function": {
+        "name": "get_weather",
+        "description": "Get the current weather for a city",
+        "parameters": {
+          "type": "object",
+          "properties": {
+            "city": {
+              "type": "string",
+              "description": "City name"
+            }
+          },
+          "required": [
+            "city"
+          ]
+        }
+      }
+    }
+  ],
+  "tool_choice": "auto",
+  "temperature": 0,
+  "max_tokens": 2048
+}
diff --git a/bench/vllm-agent/results/2026-11-08/main-0.1.94/tool.json b/bench/vllm-agent/results/2026-11-08/main-0.1.94/tool.json
new file mode 100644
index 00000000..bc893dd9
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/main-0.1.94/tool.json
@@ -0,0 +1 @@
+{"id":"chatcmpl-b0e49f1d13c90f54","object":"chat.completion","created":1794117790,"model":"Qwen3-4B","choices":[{"index":0,"message":{"role":"assistant","content":null,"refusal":null,"annotations":null,"audio":null,"function_call":null,"tool_calls":[{"id":"chatcmpl-tool-a2da8300fffa47a3","type":"function","function":{"name":"get_weather","arguments":"{\"city\": \"Paris\"}"}}],"reasoning":"\nOkay, the user is asking for the current weather in Paris. I need to use the get_weather function. Let me check the function parameters. The function requires the city name as a string. The user mentioned \"Paris,\" so I'll set the city parameter to \"Paris\". I should make sure there are no typos. Alright, the tool call should be correct.\n"},"logprobs":null,"finish_reason":"tool_calls","stop_reason":null,"token_ids":null,"routed_experts":null}],"service_tier":null,"system_fingerprint":"vllm-0.32.0-b848f2c0","usage":{"prompt_tokens":164,"total_tokens":263,"completion_tokens":99,"prompt_tokens_details":null,"completion_tokens_details":{"reasoning_tokens":76}},"prompt_logprobs":null,"prompt_token_ids":null,"prompt_text":null,"kv_transfer_params":null,"ec_transfer_params":null,"metrics":null}
\ No newline at end of file
diff --git a/bench/vllm-agent/results/2026-11-08/main-0.1.94/tools.txt b/bench/vllm-agent/results/2026-11-08/main-0.1.94/tools.txt
new file mode 100644
index 00000000..ce715234
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/main-0.1.94/tools.txt
@@ -0,0 +1,6 @@
+/home/devcloud/.local/bin/k3d
+/home/devcloud/.local/bin/kubectl
+k3d version v5.9.0
+k3s version v1.35.5-k3s1 (default)
+Client Version: v1.37.2
+Kustomize Version: v5.8.1
diff --git a/bench/vllm-agent/results/2026-11-08/main-0.1.94/up.txt b/bench/vllm-agent/results/2026-11-08/main-0.1.94/up.txt
new file mode 100644
index 00000000..50130a90
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/main-0.1.94/up.txt
@@ -0,0 +1,93 @@
+WARN: present but disabled: DSA dsa0 dsa2 dsa4, QAT 0000:01:00.0 0000:06:00.0 0000:0b:00.0 (no user work queues / VFs), so not in the cluster — enable them: k3d-gpu host-setup --apply, then k3d-gpu down && k3d-gpu up (exact commands: k3d-gpu doctor)
+==> Accelerators: intel amx
+==> Experimental features (auto): dra image-volume
+==> Pulling node image cryptoandcoffee/k3d-gpu:0.1.94...
+0.1.94: Pulling from cryptoandcoffee/k3d-gpu
+e53158a41a16: Pulling fs layer
+663ef41fb59a: Pulling fs layer
+f19171228762: Pulling fs layer
+d47d3759513d: Pulling fs layer
+44bb1521d4c8: Pulling fs layer
+dabed961d466: Pulling fs layer
+716fb9699fdf: Pulling fs layer
+69e7a4e03aa8: Pulling fs layer
+951c4ec85de1: Pulling fs layer
+f4f63f4231db: Pulling fs layer
+c6ceffcc1e6e: Pulling fs layer
+326ff02534a5: Pulling fs layer
+b9999171beec: Pulling fs layer
+aba1d122e931: Pulling fs layer
+d9f678ef87e7: Pulling fs layer
+945db06444f3: Pulling fs layer
+44136fa355b3: Already exists
+951c4ec85de1: Already exists
+c6ceffcc1e6e: Already exists
+d9f678ef87e7: Download complete
+716fb9699fdf: Download complete
+306d42c80617: Download complete
+663ef41fb59a: Download complete
+e53158a41a16: Download complete
+945db06444f3: Download complete
+d47d3759513d: Download complete
+f19171228762: Download complete
+44bb1521d4c8: Download complete
+69e7a4e03aa8: Download complete
+326ff02534a5: Download complete
+b9999171beec: Download complete
+f4f63f4231db: Download complete
+aba1d122e931: Download complete
+dabed961d466: Download complete
+dabed961d466: Pull complete
+951c4ec85de1: Pull complete
+c6ceffcc1e6e: Pull complete
+d9f678ef87e7: Pull complete
+716fb9699fdf: Pull complete
+663ef41fb59a: Pull complete
+e53158a41a16: Pull complete
+951c4ec85de1: Pull complete
+c6ceffcc1e6e: Pull complete
+d9f678ef87e7: Pull complete
+716fb9699fdf: Pull complete
+663ef41fb59a: Pull complete
+e53158a41a16: Pull complete
+945db06444f3: Pull complete
+d47d3759513d: Pull complete
+f19171228762: Pull complete
+44bb1521d4c8: Pull complete
+69e7a4e03aa8: Pull complete
+326ff02534a5: Pull complete
+b9999171beec: Pull complete
+f4f63f4231db: Pull complete
+aba1d122e931: Pull complete
+Digest: sha256:8630e14556e70111a097dcf22d15a7484bc94fa19a2d4b04b311ec60778d438e
+Status: Downloaded newer image for cryptoandcoffee/k3d-gpu:0.1.94
+docker.io/cryptoandcoffee/k3d-gpu:0.1.94
+==> Creating k3d cluster 'gpu' with image cryptoandcoffee/k3d-gpu:0.1.94...
+INFO[0000] Prep: Network                                
+INFO[0000] Created network 'k3d-gpu'                    
+INFO[0000] Created image volume k3d-gpu-images          
+INFO[0000] Starting new tools node...                   
+INFO[0000] Starting node 'k3d-gpu-tools'                
+INFO[0001] Creating node 'k3d-gpu-server-0'             
+INFO[0001] Using the k3d-tools node to gather environment information 
+INFO[0001] HostIP: using network gateway 172.19.0.1 address 
+INFO[0001] Starting cluster 'gpu'                       
+INFO[0001] Starting servers...                          
+INFO[0001] Starting node 'k3d-gpu-server-0'             
+INFO[0004] All agents already running.                  
+INFO[0004] All helpers already running.                 
+INFO[0004] Injecting records for hostAliases (incl. host.k3d.internal) and for 1 network members into CoreDNS configmap... 
+INFO[0006] Cluster 'gpu' created successfully!          
+INFO[0006] --kubeconfig-update-default=false --> sets --kubeconfig-switch-context=false 
+INFO[0006] You can now use it like this:                
+export KUBECONFIG=$(k3d kubeconfig write gpu)
+kubectl cluster-info
+==> kubectl for this cluster: export KUBECONFIG=/home/devcloud/.local/state/k3d-gpu/kubeconfig-gpu
+==> Waiting for nodes Ready...
+node/k3d-gpu-server-0 condition met
+==> Waiting for cluster DNS (CoreDNS)...
+==> intel device plugin auto-deployed by k3s.
+==> Verifying accelerators are schedulable...
+daemon set "intel-gpu-plugin" successfully rolled out
+WARN: present but disabled: DSA dsa0 dsa2 dsa4, QAT 0000:01:00.0 0000:06:00.0 0000:0b:00.0 (no user work queues / VFs), so not in the cluster — enable them: k3d-gpu host-setup --apply, then k3d-gpu down && k3d-gpu up (exact commands: k3d-gpu doctor)
+==> Accelerators schedulable: intel=4 amx=1. Try: k3d-gpu test
diff --git a/bench/vllm-agent/results/2026-11-08/main-0.1.94/vllm-previous.log b/bench/vllm-agent/results/2026-11-08/main-0.1.94/vllm-previous.log
new file mode 100644
index 00000000..4d65511b
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/main-0.1.94/vllm-previous.log
@@ -0,0 +1 @@
+Error from server (BadRequest): previous terminated container "vllm" in pod "vllm-6d68b9c67c-whhxr" not found
diff --git a/bench/vllm-agent/results/2026-11-08/main-0.1.94/vllm.log b/bench/vllm-agent/results/2026-11-08/main-0.1.94/vllm.log
new file mode 100644
index 00000000..cfd8cd7b
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/main-0.1.94/vllm.log
@@ -0,0 +1,334 @@
+W1108 06:00:50.661000 1 torch/_opaque_base.py:6] torch._opaque_base is deprecated, use torch._custom_class_base instead
+W1108 06:00:50.661000 1 torch/_library/opaque_object.py:288] register_opaque_type is deprecated, use register_custom_class instead
+W1108 06:00:50.662000 1 torch/_library/opaque_object.py:211] typ='value' is deprecated, use typ='constant' instead
+(APIServer pid=1) INFO 11-08 06:00:57 [api_utils.py:347] 
+(APIServer pid=1) INFO 11-08 06:00:57 [api_utils.py:347]        █     █     █▄   ▄█
+(APIServer pid=1) INFO 11-08 06:00:57 [api_utils.py:347]  ▄▄ ▄█ █     █     █ ▀▄▀ █  version 0.32.0
+(APIServer pid=1) INFO 11-08 06:00:57 [api_utils.py:347]   █▄█▀ █     █     █     █  model   Qwen/Qwen3-4B
+(APIServer pid=1) INFO 11-08 06:00:57 [api_utils.py:347]    ▀▀  ▀▀▀▀▀ ▀▀▀▀▀ ▀     ▀
+(APIServer pid=1) INFO 11-08 06:00:57 [api_utils.py:347] 
+(APIServer pid=1) INFO 11-08 06:00:57 [api_utils.py:286] non-default args: {'model_tag': 'Qwen/Qwen3-4B', 'enable_auto_tool_choice': True, 'tool_call_parser': 'hermes', 'model': 'Qwen/Qwen3-4B', 'dtype': 'float16', 'revision': '1cfa9a7208912126459214e8b04321603b3df60c', 'max_model_len': 8192, 'enforce_eager': True, 'served_model_name': ['Qwen3-4B'], 'reasoning_parser': 'qwen3', 'block_size': 64, 'gpu_memory_utilization': 0.9, 'enable_prefix_caching': False, 'max_num_batched_tokens': 2048}
+Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.
+(APIServer pid=1) INFO 11-08 06:01:06 [model.py:733] Resolved architecture: Qwen3ForCausalLM
+(APIServer pid=1) WARNING 11-08 06:01:06 [model.py:2498] Casting torch.bfloat16 to torch.float16.
+(APIServer pid=1) INFO 11-08 06:01:06 [model.py:2152] Using max model len 8192
+
Parse safetensors files:   0%|          | 0/3 [00:00<?, ?it/s]
Parse safetensors files:  33%|███▎      | 1/3 [00:00<00:00,  6.60it/s]
Parse safetensors files: 100%|██████████| 3/3 [00:00<00:00, 18.31it/s]
+(APIServer pid=1) WARNING 11-08 06:01:06 [vllm.py:1703] Enforce eager set, disabling torch.compile, CUDAGraphs, and JIT kernel warmup. This is equivalent to setting -cc.mode=none -cc.cudagraph_mode=none and --kernel_config.enable_jit_warmup=False
+(APIServer pid=1) WARNING 11-08 06:01:06 [vllm.py:1727] Inductor compilation was disabled by user settings, optimizations settings that are only active during inductor compilation will be ignored.
+(APIServer pid=1) INFO 11-08 06:01:06 [kernel.py:467] Final IR op priority after setting platform defaults: IrOpPriorityConfig(rms_norm=['vllm_c', 'native'], fused_add_rms_norm=['vllm_c', 'native'], gelu_and_mul_sparse=['native'])
+(APIServer pid=1) INFO 11-08 06:01:06 [vllm.py:1929] Cudagraph is disabled under eager mode
+(APIServer pid=1) INFO 11-08 06:01:06 [xpu.py:447] XPU platform: set server shutdown_timeout=5.
+
Fetching 10 files:   0%|          | 0/10 [00:00<?, ?it/s]
Fetching 10 files:  30%|███       | 3/10 [00:00<00:00, 28.76it/s]
Fetching 10 files: 100%|██████████| 10/10 [00:00<00:00, 29.68it/s]
Fetching 10 files: 100%|██████████| 10/10 [00:00<00:00, 29.50it/s]
+(APIServer pid=1) INFO 11-08 06:01:07 [compilation.py:317] Enabled custom fusions: norm_quant, act_quant
+W1108 06:01:10.474000 578 torch/_opaque_base.py:6] torch._opaque_base is deprecated, use torch._custom_class_base instead
+W1108 06:01:10.474000 578 torch/_library/opaque_object.py:288] register_opaque_type is deprecated, use register_custom_class instead
+W1108 06:01:10.475000 578 torch/_library/opaque_object.py:211] typ='value' is deprecated, use typ='constant' instead
+(EngineCore pid=578) INFO 11-08 06:01:16 [core.py:133] Initializing a V1 LLM engine (v0.32.0) with config: model='Qwen/Qwen3-4B', speculative_config=None, tokenizer='Qwen/Qwen3-4B', skip_tokenizer_init=False, tokenizer_mode=auto, revision=1cfa9a7208912126459214e8b04321603b3df60c, tokenizer_revision=1cfa9a7208912126459214e8b04321603b3df60c, trust_remote_code=False, dtype=torch.float16, max_seq_len=8192, download_dir=None, load_format=auto, tensor_parallel_size=1, pipeline_parallel_size=1, data_parallel_size=1, decode_context_parallel_size=1, dcp_comm_backend=ag_rs, disable_custom_all_reduce=True, quantization=None, quantization_config=None, enforce_eager=True, aux_output_config=AuxOutputConfig(enable_return_routed_experts=False, max_bytes=None), kv_cache_dtype=auto, device_config=xpu, structured_outputs_config=StructuredOutputsConfig(backend='auto', disable_any_whitespace=False, disable_additional_properties=False, reasoning_parser='qwen3', reasoning_parser_plugin='', enable_in_reasoning=False), observability_config=ObservabilityConfig(show_hidden_metrics_for_version=None, otlp_traces_endpoint=None, collect_detailed_traces=None, per_request_spec_decode_metrics='none', kv_cache_metrics=False, kv_cache_metrics_sample=0.01, custom_histogram_buckets=None, cudagraph_metrics=False, enable_layerwise_nvtx_tracing=False, enable_mfu_metrics=False, enable_mm_processor_stats=False, enable_logging_iteration_details=False, jit_monitor_mode='warn', jit_monitor_verbose=False), seed=0, served_model_name=Qwen3-4B, enable_prefix_caching=False, enable_chunked_prefill=True, pooler_config=None, compilation_config={'mode': <CompilationMode.NONE: 0>, 'debug_dump_path': None, 'cache_dir': '', 'compile_cache_save_format': 'binary', 'backend': 'inductor', 'custom_ops': ['all'], 'ir_enable_torch_wrap': False, 'splitting_ops': [], 'compile_mm_encoder': False, 'cudagraph_mm_encoder': False, 'encoder_cudagraph_token_budgets': [], 'encoder_cudagraph_max_vision_items_per_batch': 0, 'encoder_cudagraph_max_frames_per_batch': None, 'cudagraph_decoder_replay': True, 'decoder_replay_cudagraph_capture_sizes': [], 'decoder_replay_trim_threshold': 768, 'compile_sizes': [], 'compile_ranges_endpoints': [2048], 'inductor_compile_config': {'enable_auto_functionalized_v2': False, 'combo_kernels': True, 'benchmark_combo_kernel': True}, 'inductor_passes': {}, 'cudagraph_mode': <CUDAGraphMode.NONE: 0>, 'cudagraph_num_of_warmups': 0, 'cudagraph_capture_sizes': [], 'cudagraph_copy_inputs': False, 'cudagraph_specialize_lora': True, 'use_inductor_graph_partition': False, 'pass_config': {'fuse_norm_quant': True, 'fuse_act_quant': True, 'fuse_attn_quant': False, 'fuse_allreduce_rms': False, 'enable_qk_norm_rope_fusion': False, 'fuse_rope_kvcache_cat_mla': False, 'fuse_act_padding': False, 'fuse_qk_norm_rope_kvcache': False}, 'max_cudagraph_capture_size': 0, 'dynamic_shapes_config': {'type': <DynamicShapesType.BACKED: 'backed'>, 'evaluate_guards': False, 'assume_32_bit_indexing': False}, 'local_cache_dir': None, 'fast_moe_cold_start': False, 'static_all_moe_layers': []}, kernel_config=KernelConfig(ir_op_priority=IrOpPriorityConfig(rms_norm=['vllm_c', 'native'], fused_add_rms_norm=['vllm_c', 'native'], gelu_and_mul_sparse=['native']), enable_flashinfer_autotune=True, enable_rocm_segmented_attn_autotune=False, enable_cutedsl_warmup=True, enable_jit_warmup=False, moe_backend='auto', sparse_indexer_topk_backend='auto', linear_backend='auto', linear_backend_per_quant=None)
+(EngineCore pid=578) INFO 11-08 06:01:16 [parallel_state.py:1906] world_size=1 rank=0 local_rank=0 distributed_init_method=file:///tmp/vllm_dist_de1183cf251e471cb4f9d3e8c780165b backend=xccl
+(EngineCore pid=578) INFO 11-08 06:01:16 [parallel_state.py:2350] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, PCP rank 0, TP rank 0, ETP rank 0, EP rank N/A, EPLB rank N/A
+(EngineCore pid=578) INFO 11-08 06:01:16 [xpu_worker.py:198] Using V2 Model Runner
+(EngineCore pid=578) INFO 11-08 06:01:17 [model_runner.py:392] Loading model from scratch...
+(EngineCore pid=578) INFO 11-08 06:01:18 [xpu.py:247] Using Flash Attention backend.
+(EngineCore pid=578) INFO 11-08 06:01:18 [flash_attn.py:1117] Using FlashAttention version 2
+(EngineCore pid=578) INFO 11-08 06:01:18 [base_loader.py:82] Initializing model took 0.971832 seconds
+Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.
+(EngineCore pid=578) INFO 11-08 06:02:28 [weight_utils.py:532] Time spent downloading weights for Qwen/Qwen3-4B: 69.427395 seconds
+(EngineCore pid=578) INFO 11-08 06:02:28 [weight_utils.py:1108] Filesystem type for checkpoints: EXT4. Checkpoint size: 7.49 GiB. Available RAM: 481.56 GiB.
+(EngineCore pid=578) INFO 11-08 06:02:28 [weight_utils.py:1131] Auto-prefetch is disabled because the filesystem (EXT4) is not a recognized network FS (NFS/Lustre). If you want to force prefetching, start vLLM with --safetensors-load-strategy=prefetch.
+
Loading safetensors checkpoint shards:   0% Completed | 0/3 [00:00<?, ?it/s]
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight embed_tokens.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.0.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.0.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.0.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.0.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.0.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.0.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.1.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.1.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.1.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.1.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.1.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.1.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.10.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.10.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.10.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.10.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.10.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.10.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.11.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.11.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.11.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.11.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.11.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.11.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.12.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.12.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.12.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.12.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.12.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.12.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.13.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.13.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.13.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.13.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.13.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.13.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.14.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.14.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.14.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.14.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.14.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.14.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.15.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.15.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.15.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.2.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.2.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.2.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.2.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.2.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.2.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.3.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.3.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.3.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.3.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.3.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.3.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.4.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.4.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.4.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.4.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.4.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.4.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.5.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.5.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.5.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.5.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.5.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.5.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.6.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.6.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.6.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.6.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.6.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.6.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.7.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.7.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.7.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.7.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.7.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.7.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.8.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.8.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.8.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.8.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.8.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.8.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.9.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.9.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.9.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.9.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.9.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.9.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+
Loading safetensors checkpoint shards:  33% Completed | 1/3 [00:00<00:01,  1.55it/s]
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.15.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.15.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.15.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.16.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.16.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.16.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.16.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.16.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.16.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.17.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.17.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.17.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.17.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.17.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.17.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.18.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.18.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.18.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.18.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.18.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.18.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.19.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.19.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.19.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.19.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.19.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.19.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.20.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.20.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.20.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.20.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.20.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.20.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.21.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.21.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.21.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.21.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.21.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.21.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.22.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.22.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.22.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.22.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.22.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.22.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.23.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.23.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.23.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.23.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.23.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.23.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.24.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.24.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.24.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.24.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.24.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.24.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.25.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.25.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.25.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.25.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.25.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.25.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.26.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:28 [utils.py:333] Attempted to load weight layers.26.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.26.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.26.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.26.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.26.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.27.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.27.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.27.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.27.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.27.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.27.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.28.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.28.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.28.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.28.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.28.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.28.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.29.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.29.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.29.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.29.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.29.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.29.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.30.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.30.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.30.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.30.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.30.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.30.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.31.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.31.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.31.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.31.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.31.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.31.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.32.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.32.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.32.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.32.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.32.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.32.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.33.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.33.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.33.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.33.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.33.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.33.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.34.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.34.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.34.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.34.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.34.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.34.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.35.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.35.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.35.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+
Loading safetensors checkpoint shards:  67% Completed | 2/3 [00:01<00:00,  1.55it/s]
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.35.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.35.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight layers.35.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:02:29 [utils.py:333] Attempted to load weight norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+
Loading safetensors checkpoint shards: 100% Completed | 3/3 [00:01<00:00,  2.30it/s]
+
+(EngineCore pid=578) INFO 11-08 06:02:29 [default_loader.py:501] Loading weights took 1.31 seconds
+(EngineCore pid=578) INFO 11-08 06:02:30 [model_runner.py:424] Model loading took 7.56 GiB memory and 72.924896 seconds
+(EngineCore pid=578) INFO 11-08 06:02:30 [topk_topp_sampler.py:157] Using the fused XPU kernel for top-p & top-k sampling.
+(EngineCore pid=578) INFO 11-08 06:02:30 [torch_utils.py:295] Reducing Torch threads from 48 to 1 for serving. Set OMP_NUM_THREADS in the external environment to override.
+(EngineCore pid=578) INFO 11-08 06:02:30 [utils.py:347] Using LBNHC KV cache layout.
+(EngineCore pid=578) INFO 11-08 06:02:37 [gpu_worker.py:759] Available KV cache memory: 18.88 GiB
+(EngineCore pid=578) INFO 11-08 06:02:37 [kv_cache_utils.py:2327] XPU KV cache size: 137,472 tokens, Maximum concurrency for 8,192 tokens per request: 16.78x
+(EngineCore pid=578) INFO 11-08 06:02:37 [gpu_worker.py:1034] Free memory on device (29.92/30.3 GiB) on startup. Desired GPU memory utilization is (0.9, 27.27 GiB). Actual usage is 8.22 GiB for consumed memory (weights + non-torch), 0.17 GiB for peak activation, and 0.0 GiB for CUDAGraph memory. Replace gpu_memory_utilization config with `--kv-cache-memory=20115945882` (18.73 GiB) to fit into requested memory, or `--kv-cache-memory=22960703488` (21.38 GiB) to fully utilize gpu memory. Current kv cache memory in use is 18.88 GiB.
+(EngineCore pid=578) INFO 11-08 06:02:38 [core.py:402] init engine (profile, create kv cache, warmup model) took 7.92 s
+
Fetching 10 files:   0%|          | 0/10 [00:00<?, ?it/s]
Fetching 10 files: 100%|██████████| 10/10 [00:00<00:00, 1339.01it/s]
+
Parse safetensors files:   0%|          | 0/3 [00:00<?, ?it/s]
Parse safetensors files:  33%|███▎      | 1/3 [00:00<00:00,  7.00it/s]
Parse safetensors files: 100%|██████████| 3/3 [00:00<00:00, 18.78it/s]
+(EngineCore pid=578) WARNING 11-08 06:02:38 [vllm.py:1703] Enforce eager set, disabling torch.compile, CUDAGraphs, and JIT kernel warmup. This is equivalent to setting -cc.mode=none -cc.cudagraph_mode=none and --kernel_config.enable_jit_warmup=False
+(EngineCore pid=578) WARNING 11-08 06:02:38 [vllm.py:1727] Inductor compilation was disabled by user settings, optimizations settings that are only active during inductor compilation will be ignored.
+(EngineCore pid=578) INFO 11-08 06:02:38 [kernel.py:467] Final IR op priority after setting platform defaults: IrOpPriorityConfig(rms_norm=['vllm_c', 'native'], fused_add_rms_norm=['vllm_c', 'native'], gelu_and_mul_sparse=['native'])
+(EngineCore pid=578) INFO 11-08 06:02:38 [vllm.py:1929] Cudagraph is disabled under eager mode
+(EngineCore pid=578) INFO 11-08 06:02:38 [compilation.py:317] Enabled custom fusions: norm_quant, act_quant
+(APIServer pid=1) INFO 11-08 06:02:38 [entry.py:136] Supported tasks: ['generate']
+(APIServer pid=1) INFO 11-08 06:02:38 [factories.py:89] Scale-out endpoints are disabled. Restart vllm serve with --enable-scale-out to enable them.
+(APIServer pid=1) INFO 11-08 06:02:38 [parser_manager.py:38] "auto" tool choice has been enabled.
+(APIServer pid=1) INFO 11-08 06:02:39 [hf.py:648] Detected the chat template content format to be 'string'. You can set `--chat-template-content-format` to override this.
+(APIServer pid=1) WARNING 11-08 06:02:39 [model.py:1893] Default vLLM sampling parameters have been overridden by the model's `generation_config.json`: `{'temperature': 0.6, 'top_k': 20, 'top_p': 0.95}`. If this is not intended, please relaunch vLLM instance with `--generation-config vllm`.
+(APIServer pid=1) INFO 11-08 06:02:39 [entry.py:140] Starting vLLM server on http://0.0.0.0:8000
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:186] Available routes are:
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /openapi.json, Methods: HEAD, GET
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /docs, Methods: HEAD, GET
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /docs/oauth2-redirect, Methods: HEAD, GET
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /redoc, Methods: HEAD, GET
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /load, Methods: GET
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /version, Methods: GET
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /kv_event_sources, Methods: GET
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /health, Methods: GET
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /metrics, Methods: GET
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /tokenize, Methods: POST
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /detokenize, Methods: POST
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /v1/models, Methods: GET
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /v1/systemone, Methods: POST
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /ping, Methods: GET
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /ping, Methods: POST
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /invocations, Methods: POST
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /v1/chat/completions, Methods: POST
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /v1/chat/completions/batch, Methods: POST
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /v1/responses, Methods: POST
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /v1/responses/{response_id}, Methods: GET
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /v1/responses/{response_id}/cancel, Methods: POST
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /v1/completions, Methods: POST
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /v1/messages, Methods: POST
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /v1/messages/count_tokens, Methods: POST
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /generative_scoring, Methods: POST
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /scale_elastic_ep, Methods: POST
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /is_scaling_elastic_ep, Methods: POST
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /v1/chat/completions/render, Methods: POST
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /v1/messages/render, Methods: POST
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /v1/completions/render, Methods: POST
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /v1/responses/render, Methods: POST
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /v1/chat/completions/derender, Methods: POST
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /v1/completions/derender, Methods: POST
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /inference/v1/generate, Methods: POST
+(APIServer pid=1) INFO 11-08 06:02:39 [launcher.py:195] Route: /inference/v1/abort_requests, Methods: POST
+INFO:     Started server process [1]
+INFO:     Waiting for application startup.
+INFO:     Application startup complete.
+INFO:     10.42.0.1:52646 - "GET /health HTTP/1.1" 200 OK
+INFO:     127.0.0.1:38598 - "POST /v1/chat/completions HTTP/1.1" 200 OK
+INFO:     127.0.0.1:39446 - "GET /v1/models HTTP/1.1" 200 OK
+INFO:     10.42.0.1:53682 - "GET /health HTTP/1.1" 200 OK
+(APIServer pid=1) INFO 11-08 06:02:49 [loggers.py:335] Engine 000: Avg prompt throughput: 3.6 tokens/s, Avg generation throughput: 11.6 tokens/s, Running: 1 reqs, Waiting: 0 reqs, XPU KV cache usage: 0.1%
+INFO:     10.42.0.1:53694 - "GET /health HTTP/1.1" 200 OK
+INFO:     10.42.0.1:58598 - "GET /health HTTP/1.1" 200 OK
+(APIServer pid=1) INFO 11-08 06:02:59 [loggers.py:335] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 63.8 tokens/s, Running: 1 reqs, Waiting: 0 reqs, XPU KV cache usage: 0.6%
+INFO:     10.42.0.1:58602 - "GET /health HTTP/1.1" 200 OK
+INFO:     10.42.0.1:46442 - "GET /health HTTP/1.1" 200 OK
+(APIServer pid=1) INFO 11-08 06:03:09 [loggers.py:335] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 61.5 tokens/s, Running: 1 reqs, Waiting: 0 reqs, XPU KV cache usage: 1.0%
+INFO:     127.0.0.1:39450 - "POST /v1/chat/completions HTTP/1.1" 200 OK
+INFO:     127.0.0.1:49834 - "POST /v1/chat/completions HTTP/1.1" 200 OK
diff --git a/bench/vllm-agent/results/2026-11-08/pr357-0985b179/agent-down.txt b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/agent-down.txt
new file mode 100644
index 00000000..49cce86e
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/agent-down.txt
@@ -0,0 +1,4 @@
+deployment.apps "vllm" deleted from k3d-gpu-agent namespace
+service "vllm" deleted from k3d-gpu-agent namespace
+No resources found
+==> Model caches kept in the namespace's PVCs. Remove everything: KUBECONFIG=/home/devcloud/.local/state/k3d-gpu/kubeconfig-gpu kubectl delete namespace k3d-gpu-agent
diff --git a/bench/vllm-agent/results/2026-11-08/pr357-0985b179/agent-up.clean.txt b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/agent-up.clean.txt
new file mode 100644
index 00000000..774cacd5
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/agent-up.clean.txt
@@ -0,0 +1,14 @@
+==> agent: vllm serving Qwen/Qwen3-4B as 'Qwen3-4B' on 1x gpu.intel.com/xe, 2 CPU(s)
+namespace/k3d-gpu-agent created
+persistentvolumeclaim/vllm-models created
+deployment.apps/vllm created
+service/vllm created
+==> Waiting for the model to download and load (first run fetches it into the PVC; up to 1800s)...
+Waiting for deployment "vllm" rollout to finish: 0 of 1 updated replicas are available...
+deployment "vllm" successfully rolled out
+==> warm-up reply (2s): I am a large language model developed by Alibaba Cloud.
+==> model Qwen/Qwen3-4B at revision 1cfa9a7208912126459214e8b04321603b3df60c
+==> OpenAI API ready, model 'Qwen3-4B':
+       KUBECONFIG=/home/devcloud/.local/state/k3d-gpu/kubeconfig-gpu kubectl -n k3d-gpu-agent port-forward svc/vllm 8000:8000
+       curl http://localhost:8000/v1/chat/completions -H 'Content-Type: application/json' \
+         -d '{"model":"Qwen3-4B","messages":[{"role":"user","content":"hi"}]}'
diff --git a/bench/vllm-agent/results/2026-11-08/pr357-0985b179/agent-up.txt b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/agent-up.txt
new file mode 100644
index 00000000..d8bb73b7
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/agent-up.txt
@@ -0,0 +1,14 @@
+==> agent: vllm serving Qwen/Qwen3-4B as 'Qwen3-4B' on 1x gpu.intel.com/xe, 2 CPU(s)
+namespace/k3d-gpu-agent created
+persistentvolumeclaim/vllm-models created
+deployment.apps/vllm created
+service/vllm created
+==> Waiting for the model to download and load (first run fetches it into the PVC; up to 1800s)...
+Waiting for deployment "vllm" rollout to finish: 0 of 1 updated replicas are available...
+deployment "vllm" successfully rolled out
+==> warm-up reply (2s): I am a large language model developed by Alibaba Cloud.
+==> model Qwen/Qwen3-4B at revision 1cfa9a7208912126459214e8b04321603b3df60c
+==> OpenAI API ready, model 'Qwen3-4B':
+       KUBECONFIG=/home/devcloud/.local/state/k3d-gpu/kubeconfig-gpu kubectl -n k3d-gpu-agent port-forward svc/vllm 8000:8000
+       curl http://localhost:8000/v1/chat/completions -H 'Content-Type: application/json' \
+         -d '{"model":"Qwen3-4B","messages":[{"role":"user","content":"hi"}]}'
diff --git a/bench/vllm-agent/results/2026-11-08/pr357-0985b179/allocatable.json b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/allocatable.json
new file mode 100644
index 00000000..db5960f2
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/allocatable.json
@@ -0,0 +1 @@
+{"cpu":"96","ephemeral-storage":"3538173550480","gpu.intel.com/xe":"4","hugepages-1Gi":"0","hugepages-2Mi":"0","memory":"527767144Ki","pods":"110"}
\ No newline at end of file
diff --git a/bench/vllm-agent/results/2026-11-08/pr357-0985b179/chat-req.json b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/chat-req.json
new file mode 100644
index 00000000..d325a946
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/chat-req.json
@@ -0,0 +1,11 @@
+{
+  "model": "Qwen3-4B",
+  "messages": [
+    {
+      "role": "user",
+      "content": "What is 17 times 23?"
+    }
+  ],
+  "temperature": 0,
+  "max_tokens": 4096
+}
diff --git a/bench/vllm-agent/results/2026-11-08/pr357-0985b179/chat.json b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/chat.json
new file mode 100644
index 00000000..3183633f
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/chat.json
@@ -0,0 +1 @@
+{"id":"chatcmpl-815d5f23cec1e6a4","object":"chat.completion","created":1794119291,"model":"Qwen3-4B","choices":[{"index":0,"message":{"role":"assistant","content":"\n\nTo calculate $ 17 \\times 23 $, we can use the distributive property of multiplication:\n\n$$\n17 \\times 23 = 17 \\times (20 + 3) = (17 \\times 20) + (17 \\times 3)\n$$\n\nNow compute each part:\n\n- $ 17 \\times 20 = 340 $\n- $ 17 \\times 3 = 51 $\n\nAdd the results:\n\n$$\n340 + 51 = 391\n$$\n\nThus, the final answer is:\n\n$$\n\\boxed{391}\n$$","refusal":null,"annotations":null,"audio":null,"function_call":null,"reasoning":"\nOkay, so I need to figure out what 17 multiplied by 23 is. Hmm, let me think. I remember that multiplication can be done in different ways, like breaking it down into smaller parts or using some tricks. Let me try the standard method first.\n\nFirst, I know that 17 times 20 is 340, right? Because 17 times 10 is 170, so times 2 would be 340. Then, 17 times 3 is 51. So if I add those two together, 340 plus 51, that should be 391. Wait, is that right? Let me check again.\n\nAlternatively, maybe I can use the distributive property. 17 times 23 can be thought of as (20 + 7) times (20 + 3). But that might be more complicated. Wait, maybe another way. Let me try multiplying 17 by 23 step by step.\n\nStarting with 17 multiplied by 23. Let me write it out vertically:\n\n   17\nx23\n-----\nFirst, multiply 17 by 3 (the units place of 23). 17 times 3 is 51. Then, multiply 17 by 2 (the tens place of 23), which is 34, but since it's in the tens place, it's actually 340. Then add them together: 51 + 340 = 391. So that's the same answer as before.\n\nWait, but maybe I should verify this another way to be sure. Let me try using another method. For example, 23 times 17. Maybe I can think of 23 as 20 + 3, so 20*17 + 3*17. 20*17 is 340, and 3*17 is 51. Adding them gives 340 + 51 = 391. Same result.\n\nAlternatively, maybe using the standard multiplication algorithm. Let me write it again:\n\n   17\nx23\n-----\n   51  (17 x 3)\n+340   (17 x 20)\n-----\n  391\n\nYes, that seems correct. But maybe I can check with another approach. Let's see, 17 times 23. If I think of 23 as 25 - 2, then 17 times 25 is 425, and then subtract 17 times 2, which is 34. So 425 - 34 = 391. That's another way to get the same answer.\n\nAlternatively, using the formula (a + b)(a - b) = a² - b², but that might not apply here. Wait, 17 and 23 are not symmetric around a central number. Let me see. Maybe not helpful here.\n\nAlternatively, maybe using the fact that 17 is 10 + 7, so 10*23 + 7*23. 10*23 is 230, and 7*23 is 161. Adding them together: 230 + 161 = 391. Yep, same answer again.\n\nHmm, so all these different methods are giving me 391. That seems pretty consistent. Maybe I can check with a calculator? Wait, I don't have a calculator here, but maybe I can do a quick check with another multiplication.\n\nLet me try multiplying 23 by 17. Let's do 23*10 = 230, 23*7 = 161. 230 + 161 = 391. Yep, same result. So I think that's correct.\n\nAlternatively, maybe I can use the area model. Imagine a rectangle divided into two parts: one part is 10 by 23, and the other is 7 by 23. So the total area would be 10*23 + 7*23 = 230 + 161 = 391. That's the same as before.\n\nWait, another way: 17*23. Let me think of 17 as 15 + 2. Then 15*23 + 2*23. 15*23 is 345, and 2*23 is 46. Adding them gives 345 + 46 = 391. Again, same answer.\n\nI think I'm confident that 17 times 23 is 391. All the different methods I tried lead to the same result, so I don't think I made a mistake here. Maybe I can check with a different approach, like using the fact that 17*20 is 340, and then 17*3 is 51, so 340 + 51 is 391. Yep, that's the same as before.\n\nAlternatively, maybe using the standard multiplication steps again. Let me write it out:\n\nMultiply 17 by 3: 17*3 = 51. Then, multiply 17 by 2 (which is in the tens place), so 17*2 = 34, but since it's the tens place, it's 340. Then add 51 and 340: 51 + 340 = 391. Yep, that's the same.\n\nI think I've verified it multiple times. So I think the answer is definitely 391. I don't see any errors in my calculations. All the different methods I tried give the same result, so I'm pretty sure that's correct.\n"},"logprobs":null,"finish_reason":"stop","stop_reason":null,"token_ids":null,"routed_experts":null,"weight_versions":null}],"service_tier":null,"system_fingerprint":"vllm-0.33.0-5f414ee0","usage":{"prompt_tokens":18,"total_tokens":1444,"completion_tokens":1426,"prompt_tokens_details":null,"completion_tokens_details":{"reasoning_tokens":1283}},"prompt_logprobs":null,"prompt_token_ids":null,"prompt_text":null,"kv_transfer_params":null,"ec_transfer_params":null,"metrics":null}
\ No newline at end of file
diff --git a/bench/vllm-agent/results/2026-11-08/pr357-0985b179/deploy-vllm.yaml b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/deploy-vllm.yaml
new file mode 100644
index 00000000..370cc3b4
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/deploy-vllm.yaml
@@ -0,0 +1,156 @@
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  annotations:
+    deployment.kubernetes.io/revision: "1"
+    kubectl.kubernetes.io/last-applied-configuration: |
+      {"apiVersion":"apps/v1","kind":"Deployment","metadata":{"annotations":{},"labels":{"app":"vllm"},"name":"vllm","namespace":"k3d-gpu-agent"},"spec":{"replicas":1,"selector":{"matchLabels":{"app":"vllm"}},"strategy":{"type":"Recreate"},"template":{"metadata":{"annotations":{"k3d-gpu.io/agent-backend":"vllm","k3d-gpu.io/agent-device":"GPU"},"labels":{"app":"vllm"}},"spec":{"containers":[{"args":["Qwen/Qwen3-4B","--revision=1cfa9a7208912126459214e8b04321603b3df60c","--served-model-name=Qwen3-4B","--port=8000","--dtype=float16","--block-size=64","--max-model-len=8192","--max-num-batched-tokens=2048","--gpu-memory-utilization=0.9","--enforce-eager","--no-enable-prefix-caching","--tensor-parallel-size=1","--enable-auto-tool-choice","--tool-call-parser=hermes","--reasoning-parser=qwen3"],"env":[{"name":"VLLM_NO_USAGE_STATS","value":"1"},{"name":"DO_NOT_TRACK","value":"1"},{"name":"VLLM_WORKER_MULTIPROC_METHOD","value":"spawn"},{"name":"VLLM_ALLOW_LONG_MAX_MODEL_LEN","value":"1"},{"name":"VLLM_TARGET_DEVICE","value":"xpu"},{"name":"ZE_FLAT_DEVICE_HIERARCHY","value":"FLAT"},{"name":"ONEAPI_DEVICE_SELECTOR","value":"level_zero:*"},{"name":"ZE_ENABLE_PCI_ID_DEVICE_ORDER","value":"1"},{"name":"SYCL_CACHE_PERSISTENT","value":"1"},{"name":"ZES_ENABLE_SYSMAN","value":"1"},{"name":"HF_HUB_DISABLE_XET","value":"1"},{"name":"HF_HUB_ENABLE_HF_TRANSFER","value":"0"},{"name":"CCL_PROCESS_LAUNCHER","value":"none"},{"name":"CCL_WORKER_OFFLOAD","value":"0"},{"name":"CCL_WORKER_AFFINITY","value":""},{"name":"CCL_ZE_IPC_EXCHANGE","value":"pidfd"},{"name":"CCL_ALLREDUCE","value":"ring"},{"name":"HF_HOME","value":"/models/huggingface"},{"name":"XDG_CACHE_HOME","value":"/models/.cache"},{"name":"TRITON_CACHE_DIR","value":"/models/.triton"},{"name":"TRITON_HOME","value":"/models/.triton"}],"image":"vllm/vllm-openai-xpu:v0.33.0@sha256:be3a9aea234be64ffc2816ac561c597e0f0b1f44a61a7f4912780e4d77450086","imagePullPolicy":"IfNotPresent","name":"vllm","ports":[{"containerPort":8000,"name":"http"}],"readinessProbe":{"httpGet":{"path":"/health","port":"http"},"periodSeconds":5},"resources":{"limits":{"gpu.intel.com/xe":1},"requests":{"cpu":"2","memory":"8Gi"}},"volumeMounts":[{"mountPath":"/models","name":"models"},{"mountPath":"/dev/shm","name":"shm"}]}],"enableServiceLinks":false,"securityContext":{"supplementalGroups":[990]},"volumes":[{"name":"models","persistentVolumeClaim":{"claimName":"vllm-models"}},{"emptyDir":{"medium":"Memory","sizeLimit":"2Gi"},"name":"shm"}]}}}}
+  creationTimestamp: "2026-11-08T06:23:51Z"
+  generation: 1
+  labels:
+    app: vllm
+  name: vllm
+  namespace: k3d-gpu-agent
+  resourceVersion: "971"
+  uid: bf90b957-e133-4365-976d-f2ba3c32d729
+spec:
+  progressDeadlineSeconds: 600
+  replicas: 1
+  revisionHistoryLimit: 10
+  selector:
+    matchLabels:
+      app: vllm
+  strategy:
+    type: Recreate
+  template:
+    metadata:
+      annotations:
+        k3d-gpu.io/agent-backend: vllm
+        k3d-gpu.io/agent-device: GPU
+      labels:
+        app: vllm
+    spec:
+      containers:
+      - args:
+        - Qwen/Qwen3-4B
+        - --revision=1cfa9a7208912126459214e8b04321603b3df60c
+        - --served-model-name=Qwen3-4B
+        - --port=8000
+        - --dtype=float16
+        - --block-size=64
+        - --max-model-len=8192
+        - --max-num-batched-tokens=2048
+        - --gpu-memory-utilization=0.9
+        - --enforce-eager
+        - --no-enable-prefix-caching
+        - --tensor-parallel-size=1
+        - --enable-auto-tool-choice
+        - --tool-call-parser=hermes
+        - --reasoning-parser=qwen3
+        env:
+        - name: VLLM_NO_USAGE_STATS
+          value: "1"
+        - name: DO_NOT_TRACK
+          value: "1"
+        - name: VLLM_WORKER_MULTIPROC_METHOD
+          value: spawn
+        - name: VLLM_ALLOW_LONG_MAX_MODEL_LEN
+          value: "1"
+        - name: VLLM_TARGET_DEVICE
+          value: xpu
+        - name: ZE_FLAT_DEVICE_HIERARCHY
+          value: FLAT
+        - name: ONEAPI_DEVICE_SELECTOR
+          value: level_zero:*
+        - name: ZE_ENABLE_PCI_ID_DEVICE_ORDER
+          value: "1"
+        - name: SYCL_CACHE_PERSISTENT
+          value: "1"
+        - name: ZES_ENABLE_SYSMAN
+          value: "1"
+        - name: HF_HUB_DISABLE_XET
+          value: "1"
+        - name: HF_HUB_ENABLE_HF_TRANSFER
+          value: "0"
+        - name: CCL_PROCESS_LAUNCHER
+          value: none
+        - name: CCL_WORKER_OFFLOAD
+          value: "0"
+        - name: CCL_WORKER_AFFINITY
+        - name: CCL_ZE_IPC_EXCHANGE
+          value: pidfd
+        - name: CCL_ALLREDUCE
+          value: ring
+        - name: HF_HOME
+          value: /models/huggingface
+        - name: XDG_CACHE_HOME
+          value: /models/.cache
+        - name: TRITON_CACHE_DIR
+          value: /models/.triton
+        - name: TRITON_HOME
+          value: /models/.triton
+        image: vllm/vllm-openai-xpu:v0.33.0@sha256:be3a9aea234be64ffc2816ac561c597e0f0b1f44a61a7f4912780e4d77450086
+        imagePullPolicy: IfNotPresent
+        name: vllm
+        ports:
+        - containerPort: 8000
+          name: http
+          protocol: TCP
+        readinessProbe:
+          failureThreshold: 3
+          httpGet:
+            path: /health
+            port: http
+            scheme: HTTP
+          periodSeconds: 5
+          successThreshold: 1
+          timeoutSeconds: 1
+        resources:
+          limits:
+            gpu.intel.com/xe: "1"
+          requests:
+            cpu: "2"
+            memory: 8Gi
+        terminationMessagePath: /dev/termination-log
+        terminationMessagePolicy: File
+        volumeMounts:
+        - mountPath: /models
+          name: models
+        - mountPath: /dev/shm
+          name: shm
+      dnsPolicy: ClusterFirst
+      enableServiceLinks: false
+      restartPolicy: Always
+      schedulerName: default-scheduler
+      securityContext:
+        supplementalGroups:
+        - 990
+      terminationGracePeriodSeconds: 30
+      volumes:
+      - name: models
+        persistentVolumeClaim:
+          claimName: vllm-models
+      - emptyDir:
+          medium: Memory
+          sizeLimit: 2Gi
+        name: shm
+status:
+  availableReplicas: 1
+  conditions:
+  - lastTransitionTime: "2026-11-08T06:28:06Z"
+    lastUpdateTime: "2026-11-08T06:28:06Z"
+    message: Deployment has minimum availability.
+    reason: MinimumReplicasAvailable
+    status: "True"
+    type: Available
+  - lastTransitionTime: "2026-11-08T06:23:51Z"
+    lastUpdateTime: "2026-11-08T06:28:06Z"
+    message: ReplicaSet "vllm-78c46bdf8f" has successfully progressed.
+    reason: NewReplicaSetAvailable
+    status: "True"
+    type: Progressing
+  observedGeneration: 1
+  readyReplicas: 1
+  replicas: 1
+  terminatingReplicas: 0
+  updatedReplicas: 1
diff --git a/bench/vllm-agent/results/2026-11-08/pr357-0985b179/describe-pods.txt b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/describe-pods.txt
new file mode 100644
index 00000000..2d961155
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/describe-pods.txt
@@ -0,0 +1,110 @@
+Name:             vllm-78c46bdf8f-668sw
+Namespace:        k3d-gpu-agent
+Priority:         0
+Service Account:  default
+Node:             k3d-gpu-server-0/172.19.0.2
+Start Time:       Sun, 08 Nov 2026 06:23:56 +0000
+Labels:           app=vllm
+                  pod-template-hash=78c46bdf8f
+Annotations:      k3d-gpu.io/agent-backend: vllm
+                  k3d-gpu.io/agent-device: GPU
+Status:           Running
+IP:               10.42.0.11
+IPs:
+  IP:           10.42.0.11
+Controlled By:  ReplicaSet/vllm-78c46bdf8f
+Containers:
+  vllm:
+    Container ID:  containerd://a758b048e62e0517492fbecaade67314e6f376808fa2c8c06998fc1a3cb16214
+    Image:         vllm/vllm-openai-xpu:v0.33.0@sha256:be3a9aea234be64ffc2816ac561c597e0f0b1f44a61a7f4912780e4d77450086
+    Image ID:      docker.io/vllm/vllm-openai-xpu@sha256:be3a9aea234be64ffc2816ac561c597e0f0b1f44a61a7f4912780e4d77450086
+    Port:          8000/TCP (http)
+    Host Port:     0/TCP (http)
+    Args:
+      Qwen/Qwen3-4B
+      --revision=1cfa9a7208912126459214e8b04321603b3df60c
+      --served-model-name=Qwen3-4B
+      --port=8000
+      --dtype=float16
+      --block-size=64
+      --max-model-len=8192
+      --max-num-batched-tokens=2048
+      --gpu-memory-utilization=0.9
+      --enforce-eager
+      --no-enable-prefix-caching
+      --tensor-parallel-size=1
+      --enable-auto-tool-choice
+      --tool-call-parser=hermes
+      --reasoning-parser=qwen3
+    State:          Running
+      Started:      Sun, 08 Nov 2026 06:26:10 +0000
+    Ready:          True
+    Restart Count:  0
+    Limits:
+      gpu.intel.com/xe:  1
+    Requests:
+      cpu:               2
+      gpu.intel.com/xe:  1
+      memory:            8Gi
+    Readiness:           http-get http://:http/health delay=0s timeout=1s period=5s successThreshold=1 failureThreshold=3
+    Environment:
+      VLLM_NO_USAGE_STATS:            1
+      DO_NOT_TRACK:                   1
+      VLLM_WORKER_MULTIPROC_METHOD:   spawn
+      VLLM_ALLOW_LONG_MAX_MODEL_LEN:  1
+      VLLM_TARGET_DEVICE:             xpu
+      ZE_FLAT_DEVICE_HIERARCHY:       FLAT
+      ONEAPI_DEVICE_SELECTOR:         level_zero:*
+      ZE_ENABLE_PCI_ID_DEVICE_ORDER:  1
+      SYCL_CACHE_PERSISTENT:          1
+      ZES_ENABLE_SYSMAN:              1
+      HF_HUB_DISABLE_XET:             1
+      HF_HUB_ENABLE_HF_TRANSFER:      0
+      CCL_PROCESS_LAUNCHER:           none
+      CCL_WORKER_OFFLOAD:             0
+      CCL_WORKER_AFFINITY:            
+      CCL_ZE_IPC_EXCHANGE:            pidfd
+      CCL_ALLREDUCE:                  ring
+      HF_HOME:                        /models/huggingface
+      XDG_CACHE_HOME:                 /models/.cache
+      TRITON_CACHE_DIR:               /models/.triton
+      TRITON_HOME:                    /models/.triton
+    Mounts:
+      /dev/shm from shm (rw)
+      /models from models (rw)
+      /var/run/secrets/kubernetes.io/serviceaccount from kube-api-access-8wrzs (ro)
+Conditions:
+  Type                        Status
+  PodReadyToStartContainers   True 
+  Initialized                 True 
+  Ready                       True 
+  ContainersReady             True 
+  PodScheduled                True 
+Volumes:
+  models:
+    Type:       PersistentVolumeClaim (a reference to a PersistentVolumeClaim in the same namespace)
+    ClaimName:  vllm-models
+    ReadOnly:   false
+  shm:
+    Type:       EmptyDir (a temporary directory that shares a pod's lifetime)
+    Medium:     Memory
+    SizeLimit:  2Gi
+  kube-api-access-8wrzs:
+    Type:                    Projected (a volume that contains injected data from multiple sources)
+    TokenExpirationSeconds:  3607
+    ConfigMapName:           kube-root-ca.crt
+    Optional:                false
+    DownwardAPI:             true
+QoS Class:                   Burstable
+Node-Selectors:              <none>
+Tolerations:                 node.kubernetes.io/not-ready:NoExecute op=Exists for 300s
+                             node.kubernetes.io/unreachable:NoExecute op=Exists for 300s
+Events:
+  Type     Reason     Age                 From               Message
+  ----     ------     ----                ----               -------
+  Normal   Scheduled  4m12s               default-scheduler  Successfully assigned k3d-gpu-agent/vllm-78c46bdf8f-668sw to k3d-gpu-server-0
+  Normal   Pulling    4m12s               kubelet            spec.containers{vllm}: Pulling image "vllm/vllm-openai-xpu:v0.33.0@sha256:be3a9aea234be64ffc2816ac561c597e0f0b1f44a61a7f4912780e4d77450086"
+  Normal   Pulled     118s                kubelet            spec.containers{vllm}: Successfully pulled image "vllm/vllm-openai-xpu:v0.33.0@sha256:be3a9aea234be64ffc2816ac561c597e0f0b1f44a61a7f4912780e4d77450086" in 2m13.447s (2m13.447s including waiting). Image size: 5212361514 bytes.
+  Normal   Created    118s                kubelet            spec.containers{vllm}: Container created
+  Normal   Started    118s                kubelet            spec.containers{vllm}: Container started
+  Warning  Unhealthy  7s (x25 over 118s)  kubelet            spec.containers{vllm}: Readiness probe failed: Get "http://10.42.0.11:8000/health": dial tcp 10.42.0.11:8000: connect: connection refused
diff --git a/bench/vllm-agent/results/2026-11-08/pr357-0985b179/down.txt b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/down.txt
new file mode 100644
index 00000000..06ca520f
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/down.txt
@@ -0,0 +1,7 @@
+==> Deleting k3d cluster 'gpu'...
+INFO[0000] Deleting cluster 'gpu'                       
+INFO[0003] Deleting cluster network 'k3d-gpu'           
+INFO[0003] Deleting 1 attached volumes...               
+INFO[0003] Removing cluster details from default kubeconfig... 
+INFO[0003] Removing standalone kubeconfig file (if there is one)... 
+INFO[0003] Successfully deleted cluster gpu!            
diff --git a/bench/vllm-agent/results/2026-11-08/pr357-0985b179/driver.log b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/driver.log
new file mode 100644
index 00000000..de1b1f1e
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/driver.log
@@ -0,0 +1,25 @@
+2026-11-08T06:23:01Z [pr357-0985b179] launcher /home/devcloud/k3dgpu-bench/k3d-gpu-pr357-0985b179/scripts/k3d-gpu: k3d-gpu 0.1.95
+2026-11-08T06:23:01Z [pr357-0985b179] vram pre: 0000:12:00.0 card1 0.00;0000:48:00.0 card2 0.00;0000:7c:00.0 card3 0.00;0000:96:00.0 card4 0.00;
+2026-11-08T06:23:50Z [pr357-0985b179] up rc=0 in 49 s
+2026-11-08T06:23:50Z [pr357-0985b179] kubeconfig /home/devcloud/.local/state/k3d-gpu/kubeconfig-gpu
+2026-11-08T06:23:50Z [pr357-0985b179] allocatable {"cpu":"96","ephemeral-storage":"3538173550480","gpu.intel.com/xe":"4","hugepages-1Gi":"0","hugepages-2Mi":"0","memory":"527767144Ki","pods":"110"}
+2026-11-08T06:28:08Z [pr357-0985b179] agent up rc=0 in 258 s
+2026-11-08T06:28:08Z [pr357-0985b179] launcher warm-up line: warm-up reply (2s): I am a large language model developed by Alibaba Cloud.
+2026-11-08T06:28:08Z [pr357-0985b179] pod Ready=True limits={"gpu.intel.com/xe":"1"}
+2026-11-08T06:28:08Z [pr357-0985b179] pod image vllm/vllm-openai-xpu:v0.33.0@sha256:be3a9aea234be64ffc2816ac561c597e0f0b1f44a61a7f4912780e4d77450086 docker.io/vllm/vllm-openai-xpu@sha256:be3a9aea234be64ffc2816ac561c597e0f0b1f44a61a7f4912780e4d77450086 restarts=0
+2026-11-08T06:28:08Z [pr357-0985b179] pod enableServiceLinks=false
+2026-11-08T06:28:08Z [pr357-0985b179] pod render nodes: card1 renderD128 
+2026-11-08T06:28:08Z [pr357-0985b179] pod VLLM_PORT lines: 0
+2026-11-08T06:28:08Z [pr357-0985b179] vram serving: 0000:12:00.0 card1 27.30;0000:48:00.0 card2 0.00;0000:7c:00.0 card3 0.00;0000:96:00.0 card4 0.00;
+2026-11-08T06:28:08Z [pr357-0985b179] endpoint command: KUBECONFIG=/home/devcloud/.local/state/k3d-gpu/kubeconfig-gpu kubectl -n k3d-gpu-agent port-forward svc/vllm 8000:8000
+2026-11-08T06:28:11Z [pr357-0985b179] models: ["Qwen3-4B"]
+2026-11-08T06:28:33Z [pr357-0985b179] chat http=200 total_s=22.721185 finish=stop usage={"prompt_tokens":18,"total_tokens":1444,"completion_tokens":1426,"prompt_tokens_details":null,"completion_tokens_details":{"reasoning_tokens":1283}}
+2026-11-08T06:28:33Z [pr357-0985b179] chat content:   To calculate $ 17 \times 23 $, we can use the distributive property of multiplication:  $$ 17 \times 23 = 17 \times (20 + 3) = (17 \times 20) + (17 \times 3) $$  Now compute each part:  - $ 17 \times 20 = 340 $ - $ 17 \times 3 = 51 $  Add the results:  $$ 340 + 51 = 391 $$  Thus, the final answer is:  $$ \boxed{391} $$ 
+2026-11-08T06:28:33Z [pr357-0985b179] chat reasoning chars: 3593
+2026-11-08T06:28:35Z [pr357-0985b179] tool http=200 total_s=1.623996 finish=tool_calls tool_calls=[{"id":"chatcmpl-tool-8874486735506104","type":"function","function":{"name":"get_weather","arguments":"{\"city\": \"Paris\"}"}}]
+2026-11-08T06:28:35Z [pr357-0985b179] tool arguments parse: {"city":"Paris"}
+2026-11-08T06:28:35Z [pr357-0985b179] agent down rc=0
+2026-11-08T06:28:39Z [pr357-0985b179] down rc=0
+2026-11-08T06:28:44Z [pr357-0985b179] vram after: 0000:12:00.0 card1 0.00;0000:48:00.0 card2 0.00;0000:7c:00.0 card3 0.00;0000:96:00.0 card4 0.00;
+2026-11-08T06:28:44Z [pr357-0985b179] k3d containers after: 0
+2026-11-08T06:28:44Z [pr357-0985b179] RESULT PASS
diff --git a/bench/vllm-agent/results/2026-11-08/pr357-0985b179/events.txt b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/events.txt
new file mode 100644
index 00000000..aa88cd9c
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/events.txt
@@ -0,0 +1,146 @@
+NAMESPACE                LAST SEEN   TYPE      REASON                           OBJECT                                                           MESSAGE
+kube-system              5m11s       Normal    Scheduled                        pod/helm-install-gateway-api-crd-8zw5k                           Successfully assigned kube-system/helm-install-gateway-api-crd-8zw5k to k3d-gpu-server-0
+node-feature-discovery   5m11s       Normal    Scheduled                        pod/nfd-master-74b44b767d-b4lcv                                  Successfully assigned node-feature-discovery/nfd-master-74b44b767d-b4lcv to k3d-gpu-server-0
+kube-system              5m11s       Normal    Scheduled                        pod/intel-gpu-plugin-4nfj2                                       Successfully assigned kube-system/intel-gpu-plugin-4nfj2 to k3d-gpu-server-0
+kube-system              5m11s       Normal    Scheduled                        pod/coredns-577d995dff-m4bm5                                     Successfully assigned kube-system/coredns-577d995dff-m4bm5 to k3d-gpu-server-0
+kube-system              5m11s       Normal    Scheduled                        pod/metrics-server-778c4649b4-rl6zq                              Successfully assigned kube-system/metrics-server-778c4649b4-rl6zq to k3d-gpu-server-0
+kube-system              5m11s       Normal    Scheduled                        pod/local-path-provisioner-6858d854cf-fjp42                      Successfully assigned kube-system/local-path-provisioner-6858d854cf-fjp42 to k3d-gpu-server-0
+k3d-gpu-agent            4m39s       Normal    Scheduled                        pod/vllm-78c46bdf8f-668sw                                        Successfully assigned k3d-gpu-agent/vllm-78c46bdf8f-668sw to k3d-gpu-server-0
+node-feature-discovery   5m11s       Normal    Scheduled                        pod/nfd-gc-8d4595cbc-82fbz                                       Successfully assigned node-feature-discovery/nfd-gc-8d4595cbc-82fbz to k3d-gpu-server-0
+default                  5m16s       Normal    Starting                         node/k3d-gpu-server-0                                            
+node-feature-discovery   5m11s       Normal    Scheduled                        pod/nfd-worker-2fbwn                                             Successfully assigned node-feature-discovery/nfd-worker-2fbwn to k3d-gpu-server-0
+default                  5m18s       Normal    CertificateExpirationOK          node/k3d-gpu-server-0                                            Node and Certificate Authority certificates managed by k3s are OK
+default                  5m18s       Normal    NodeAllocatableEnforced          node/k3d-gpu-server-0                                            Updated Node Allocatable limit across pods
+default                  5m18s       Normal    Starting                         node/k3d-gpu-server-0                                            Starting kubelet.
+default                  5m18s       Warning   InvalidDiskCapacity              node/k3d-gpu-server-0                                            invalid capacity 0 on image filesystem
+default                  5m17s       Normal    NodeReady                        node/k3d-gpu-server-0                                            Node k3d-gpu-server-0 status is now: NodeReady
+default                  5m17s       Normal    NodeHasSufficientPID             node/k3d-gpu-server-0                                            Node k3d-gpu-server-0 status is now: NodeHasSufficientPID
+default                  5m17s       Normal    NodeHasNoDiskPressure            node/k3d-gpu-server-0                                            Node k3d-gpu-server-0 status is now: NodeHasNoDiskPressure
+default                  5m17s       Normal    NodeHasSufficientMemory          node/k3d-gpu-server-0                                            Node k3d-gpu-server-0 status is now: NodeHasSufficientMemory
+kube-system              5m16s       Normal    ApplyingManifest                 addon/intel-dsa-plugin                                           Applying manifest at "/var/lib/rancher/k3s/server/manifests/intel-dsa-plugin.yaml"
+kube-system              5m16s       Normal    AppliedManifest                  addon/intel-dsa-plugin                                           Applied manifest at "/var/lib/rancher/k3s/server/manifests/intel-dsa-plugin.yaml"
+kube-system              5m16s       Normal    DeletingManifest                 addon/intel-gpu-dra                                              Deleting manifest at "/var/lib/rancher/k3s/server/manifests/intel-gpu-dra.yaml"
+kube-system              5m16s       Normal    ApplyingManifest                 addon/ccm                                                        Applying manifest at "/var/lib/rancher/k3s/server/manifests/ccm.yaml"
+kube-system              5m16s       Normal    ApplyingManifest                 addon/intel-gpu-plugin                                           Applying manifest at "/var/lib/rancher/k3s/server/manifests/intel-gpu-plugin.yaml"
+kube-system              5m16s       Normal    ApplyingManifest                 addon/gaudi-device-plugin                                        Applying manifest at "/var/lib/rancher/k3s/server/manifests/gaudi-device-plugin.yaml"
+kube-system              5m16s       Normal    AppliedManifest                  addon/gateway-api-crd                                            Applied manifest at "/var/lib/rancher/k3s/server/manifests/gateway-api-crd.yaml"
+kube-system              5m16s       Normal    ApplyingManifest                 addon/gateway-api-crd                                            Applying manifest at "/var/lib/rancher/k3s/server/manifests/gateway-api-crd.yaml"
+kube-system              5m16s       Normal    AppliedManifest                  addon/coredns                                                    Applied manifest at "/var/lib/rancher/k3s/server/manifests/coredns.yaml"
+kube-system              5m16s       Normal    ApplyingManifest                 addon/coredns                                                    Applying manifest at "/var/lib/rancher/k3s/server/manifests/coredns.yaml"
+kube-system              5m16s       Normal    AppliedManifest                  addon/gaudi-device-plugin                                        Applied manifest at "/var/lib/rancher/k3s/server/manifests/gaudi-device-plugin.yaml"
+kube-system              5m16s       Normal    AppliedManifest                  addon/ccm                                                        Applied manifest at "/var/lib/rancher/k3s/server/manifests/ccm.yaml"
+kube-system              5m15s       Normal    ApplyingManifest                 addon/node-feature-discovery                                     Applying manifest at "/var/lib/rancher/k3s/server/manifests/node-feature-discovery.yaml"
+kube-system              5m15s       Normal    AppliedManifest                  addon/auth-reader                                                Applied manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/auth-reader.yaml"
+kube-system              5m15s       Normal    ApplyingManifest                 addon/auth-reader                                                Applying manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/auth-reader.yaml"
+kube-system              5m15s       Normal    ApplyingManifest                 addon/metrics-server-deployment                                  Applying manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/metrics-server-deployment.yaml"
+kube-system              5m15s       Normal    AppliedManifest                  addon/metrics-server-deployment                                  Applied manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/metrics-server-deployment.yaml"
+kube-system              5m15s       Normal    ApplyingManifest                 addon/metrics-server-service                                     Applying manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/metrics-server-service.yaml"
+kube-system              5m15s       Normal    AppliedManifest                  addon/metrics-server-service                                     Applied manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/metrics-server-service.yaml"
+default                  5m15s       Normal    Synced                           node/k3d-gpu-server-0                                            Node synced successfully
+kube-system              5m15s       Normal    AppliedManifest                  addon/intel-gpu-plugin                                           Applied manifest at "/var/lib/rancher/k3s/server/manifests/intel-gpu-plugin.yaml"
+kube-system              5m15s       Normal    AppliedManifest                  addon/metrics-apiservice                                         Applied manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/metrics-apiservice.yaml"
+kube-system              5m15s       Normal    AppliedManifest                  addon/auth-delegator                                             Applied manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/auth-delegator.yaml"
+kube-system              5m15s       Normal    ApplyingManifest                 addon/auth-delegator                                             Applying manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/auth-delegator.yaml"
+kube-system              5m15s       Normal    ApplyingManifest                 addon/metrics-apiservice                                         Applying manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/metrics-apiservice.yaml"
+kube-system              5m15s       Normal    AppliedManifest                  addon/aggregated-metrics-reader                                  Applied manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/aggregated-metrics-reader.yaml"
+kube-system              5m15s       Normal    ApplyingManifest                 addon/aggregated-metrics-reader                                  Applying manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/aggregated-metrics-reader.yaml"
+kube-system              5m15s       Normal    ApplyJob                         helmchart/gateway-api-crd                                        Applying HelmChart from https://%{KUBERNETES_API}%/static/charts/gateway-api-crd-1.6.103.tgz using Job kube-system/helm-install-gateway-api-crd
+kube-system              5m15s       Normal    AppliedManifest                  addon/local-storage                                              Applied manifest at "/var/lib/rancher/k3s/server/manifests/local-storage.yaml"
+kube-system              5m15s       Normal    ApplyingManifest                 addon/local-storage                                              Applying manifest at "/var/lib/rancher/k3s/server/manifests/local-storage.yaml"
+kube-system              5m15s       Normal    AppliedManifest                  addon/resource-reader                                            Applied manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/resource-reader.yaml"
+kube-system              5m15s       Normal    ApplyingManifest                 addon/resource-reader                                            Applying manifest at "/var/lib/rancher/k3s/server/manifests/metrics-server/resource-reader.yaml"
+kube-system              5m15s       Normal    ApplyingManifest                 addon/intel-gpu-wsl-plugin                                       Applying manifest at "/var/lib/rancher/k3s/server/manifests/intel-gpu-wsl-plugin.yaml"
+kube-system              5m15s       Normal    AppliedManifest                  addon/intel-gpu-wsl-plugin                                       Applied manifest at "/var/lib/rancher/k3s/server/manifests/intel-gpu-wsl-plugin.yaml"
+default                  5m15s       Normal    NodePasswordValidationComplete   node/k3d-gpu-server-0                                            Deferred node password secret validation complete
+kube-system              5m15s       Normal    AppliedManifest                  addon/intel-qat-plugin                                           Applied manifest at "/var/lib/rancher/k3s/server/manifests/intel-qat-plugin.yaml"
+kube-system              5m15s       Normal    ApplyingManifest                 addon/intel-qat-plugin                                           Applying manifest at "/var/lib/rancher/k3s/server/manifests/intel-qat-plugin.yaml"
+kube-system              5m15s       Normal    ApplyingManifest                 addon/intel-iaa-plugin                                           Applying manifest at "/var/lib/rancher/k3s/server/manifests/intel-iaa-plugin.yaml"
+kube-system              5m15s       Normal    AppliedManifest                  addon/intel-npu-plugin                                           Applied manifest at "/var/lib/rancher/k3s/server/manifests/intel-npu-plugin.yaml"
+kube-system              5m15s       Normal    ApplyingManifest                 addon/intel-npu-plugin                                           Applying manifest at "/var/lib/rancher/k3s/server/manifests/intel-npu-plugin.yaml"
+kube-system              5m15s       Normal    AppliedManifest                  addon/intel-iaa-plugin                                           Applied manifest at "/var/lib/rancher/k3s/server/manifests/intel-iaa-plugin.yaml"
+kube-system              5m12s       Normal    Suspended                        job/helm-install-gateway-api-crd                                 Job suspended
+default                  5m12s       Normal    RegisteredNode                   node/k3d-gpu-server-0                                            Node k3d-gpu-server-0 event: Registered Node k3d-gpu-server-0 in Controller
+kube-system              5m12s       Normal    AppliedManifest                  addon/nvidia-device-plugin                                       Applied manifest at "/var/lib/rancher/k3s/server/manifests/nvidia-device-plugin.yaml"
+kube-system              5m12s       Normal    ApplyingManifest                 addon/nvidia-device-plugin                                       Applying manifest at "/var/lib/rancher/k3s/server/manifests/nvidia-device-plugin.yaml"
+kube-system              5m12s       Normal    ApplyingManifest                 addon/runtimes                                                   Applying manifest at "/var/lib/rancher/k3s/server/manifests/runtimes.yaml"
+kube-system              5m12s       Normal    AppliedManifest                  addon/rolebindings                                               Applied manifest at "/var/lib/rancher/k3s/server/manifests/rolebindings.yaml"
+kube-system              5m12s       Normal    ApplyingManifest                 addon/rolebindings                                               Applying manifest at "/var/lib/rancher/k3s/server/manifests/rolebindings.yaml"
+kube-system              5m12s       Normal    DeletingManifest                 addon/nri-balloons                                               Deleting manifest at "/var/lib/rancher/k3s/server/manifests/nri-balloons.yaml"
+kube-system              5m12s       Normal    AppliedManifest                  addon/node-feature-discovery                                     Applied manifest at "/var/lib/rancher/k3s/server/manifests/node-feature-discovery.yaml"
+kube-system              5m11s       Warning   FailedCreatePodSandBox           pod/intel-gpu-plugin-4nfj2                                       Failed to create pod sandbox: rpc error: code = Unknown desc = failed to setup network for sandbox "05b039da629eaf48bc8cda20a7740b84b437a81b1071c2909a3a89b3d037f816": plugin type="flannel" failed (add): failed to load flannel 'subnet.env' file: open /run/flannel/subnet.env: no such file or directory. Check the flannel pod log for this node.
+kube-system              5m11s       Normal    ScalingReplicaSet                deployment/coredns                                               Scaled up replica set coredns-577d995dff from 0 to 1
+kube-system              5m11s       Warning   FailedCreatePodSandBox           pod/coredns-577d995dff-m4bm5                                     Failed to create pod sandbox: rpc error: code = Unknown desc = failed to setup network for sandbox "afda2d90ec752371f39e205d566ed000d3e7c16eb3ead9d9fdc7f8a583a15354": plugin type="flannel" failed (add): failed to load flannel 'subnet.env' file: open /run/flannel/subnet.env: no such file or directory. Check the flannel pod log for this node.
+node-feature-discovery   5m11s       Normal    SuccessfulCreate                 daemonset/nfd-worker                                             Created pod: nfd-worker-2fbwn
+node-feature-discovery   5m11s       Warning   FailedCreatePodSandBox           pod/nfd-worker-2fbwn                                             Failed to create pod sandbox: rpc error: code = Unknown desc = failed to setup network for sandbox "aef81fc07be669c93ba75f278b6f872ab8f82eed2590547de85a0df707048a5a": plugin type="flannel" failed (add): failed to load flannel 'subnet.env' file: open /run/flannel/subnet.env: no such file or directory. Check the flannel pod log for this node.
+kube-system              5m11s       Normal    SuccessfulCreate                 replicaset/metrics-server-778c4649b4                             Created pod: metrics-server-778c4649b4-rl6zq
+kube-system              5m11s       Normal    SuccessfulCreate                 daemonset/intel-gpu-plugin                                       Created pod: intel-gpu-plugin-4nfj2
+kube-system              5m11s       Normal    AppliedManifest                  addon/runtimes                                                   Applied manifest at "/var/lib/rancher/k3s/server/manifests/runtimes.yaml"
+kube-system              5m11s       Normal    DeletingManifest                 addon/traefik                                                    Deleting manifest at "/var/lib/rancher/k3s/server/manifests/traefik.yaml"
+node-feature-discovery   5m11s       Warning   FailedCreatePodSandBox           pod/nfd-gc-8d4595cbc-82fbz                                       Failed to create pod sandbox: rpc error: code = Unknown desc = failed to setup network for sandbox "3f7ba6b9af137145234fe891f127f54e5fb0687f549bfd2f8c4c108b6200c0c2": plugin type="flannel" failed (add): failed to load flannel 'subnet.env' file: open /run/flannel/subnet.env: no such file or directory. Check the flannel pod log for this node.
+kube-system              5m11s       Normal    ScalingReplicaSet                deployment/metrics-server                                        Scaled up replica set metrics-server-778c4649b4 from 0 to 1
+kube-system              5m11s       Normal    Resumed                          job/helm-install-gateway-api-crd                                 Job resumed
+kube-system              5m11s       Normal    SuccessfulCreate                 job/helm-install-gateway-api-crd                                 Created pod: helm-install-gateway-api-crd-8zw5k
+node-feature-discovery   5m11s       Normal    SuccessfulCreate                 replicaset/nfd-gc-8d4595cbc                                      Created pod: nfd-gc-8d4595cbc-82fbz
+node-feature-discovery   5m11s       Normal    ScalingReplicaSet                deployment/nfd-gc                                                Scaled up replica set nfd-gc-8d4595cbc from 0 to 1
+kube-system              5m11s       Warning   FailedCreatePodSandBox           pod/metrics-server-778c4649b4-rl6zq                              Failed to create pod sandbox: rpc error: code = Unknown desc = failed to setup network for sandbox "aaeda7117447c5580a05cddf25a2d0121d952acaf59dacf0c4e5ec3592c2260e": plugin type="flannel" failed (add): failed to load flannel 'subnet.env' file: open /run/flannel/subnet.env: no such file or directory. Check the flannel pod log for this node.
+kube-system              5m11s       Warning   FailedCreatePodSandBox           pod/local-path-provisioner-6858d854cf-fjp42                      Failed to create pod sandbox: rpc error: code = Unknown desc = failed to setup network for sandbox "454593e462bf63e7ac9004cb6a295df3de056bad67d91d57c57fb95282d9817d": plugin type="flannel" failed (add): failed to load flannel 'subnet.env' file: open /run/flannel/subnet.env: no such file or directory. Check the flannel pod log for this node.
+node-feature-discovery   5m11s       Normal    ScalingReplicaSet                deployment/nfd-master                                            Scaled up replica set nfd-master-74b44b767d from 0 to 1
+kube-system              5m11s       Warning   FailedCreatePodSandBox           pod/helm-install-gateway-api-crd-8zw5k                           Failed to create pod sandbox: rpc error: code = Unknown desc = failed to setup network for sandbox "4386216f68b8bce4afae1ac3c36c5b9caab1caaf743688ee07d2f554ce824495": plugin type="flannel" failed (add): failed to load flannel 'subnet.env' file: open /run/flannel/subnet.env: no such file or directory. Check the flannel pod log for this node.
+node-feature-discovery   5m11s       Warning   FailedCreatePodSandBox           pod/nfd-master-74b44b767d-b4lcv                                  Failed to create pod sandbox: rpc error: code = Unknown desc = failed to setup network for sandbox "557bcb70f275a30fc9f308517f01164f244a1d9f7b49682590e33580f9291f05": plugin type="flannel" failed (add): failed to load flannel 'subnet.env' file: open /run/flannel/subnet.env: no such file or directory. Check the flannel pod log for this node.
+kube-system              5m11s       Normal    SuccessfulCreate                 replicaset/coredns-577d995dff                                    Created pod: coredns-577d995dff-m4bm5
+kube-system              5m11s       Normal    SuccessfulCreate                 replicaset/local-path-provisioner-6858d854cf                     Created pod: local-path-provisioner-6858d854cf-fjp42
+kube-system              5m11s       Normal    ScalingReplicaSet                deployment/local-path-provisioner                                Scaled up replica set local-path-provisioner-6858d854cf from 0 to 1
+node-feature-discovery   5m11s       Normal    SuccessfulCreate                 replicaset/nfd-master-74b44b767d                                 Created pod: nfd-master-74b44b767d-b4lcv
+kube-system              5m11s       Normal    ResumeJob                        helmchart/gateway-api-crd                                        Resumed synced Job kube-system/helm-install-gateway-api-crd
+node-feature-discovery   4m58s       Normal    Pulling                          pod/nfd-master-74b44b767d-b4lcv                                  Pulling image "registry.k8s.io/nfd/node-feature-discovery:v0.20.1@sha256:6fc5c9af9176e1acb5ce57a4a1d0c0d3eb25cd34f5bdd0efefebf0efe45a8379"
+kube-system              4m58s       Normal    Pulling                          pod/local-path-provisioner-6858d854cf-fjp42                      Pulling image "rancher/local-path-provisioner:v0.0.37"
+node-feature-discovery   4m58s       Normal    Pulling                          pod/nfd-gc-8d4595cbc-82fbz                                       Pulling image "registry.k8s.io/nfd/node-feature-discovery:v0.20.1@sha256:6fc5c9af9176e1acb5ce57a4a1d0c0d3eb25cd34f5bdd0efefebf0efe45a8379"
+kube-system              4m58s       Normal    Pulling                          pod/intel-gpu-plugin-4nfj2                                       Pulling image "intel/intel-gpu-plugin:0.37.1@sha256:004641b2547d58f1eadce58473fa160ad71377de1eb588cf1f82832387bf3e33"
+node-feature-discovery   4m58s       Normal    Pulling                          pod/nfd-worker-2fbwn                                             Pulling image "registry.k8s.io/nfd/node-feature-discovery:v0.20.1@sha256:6fc5c9af9176e1acb5ce57a4a1d0c0d3eb25cd34f5bdd0efefebf0efe45a8379"
+kube-system              4m56s       Normal    Pulled                           pod/intel-gpu-plugin-4nfj2                                       Successfully pulled image "intel/intel-gpu-plugin:0.37.1@sha256:004641b2547d58f1eadce58473fa160ad71377de1eb588cf1f82832387bf3e33" in 1.994s (1.994s including waiting). Image size: 5970136 bytes.
+kube-system              4m56s       Normal    Pulling                          pod/helm-install-gateway-api-crd-8zw5k                           Pulling image "rancher/klipper-helm:v0.13.3-build20260727"
+kube-system              4m56s       Normal    Pulling                          pod/metrics-server-778c4649b4-rl6zq                              Pulling image "rancher/mirrored-metrics-server:v0.9.0"
+kube-system              4m56s       Normal    Started                          pod/intel-gpu-plugin-4nfj2                                       Container started
+kube-system              4m56s       Normal    Created                          pod/intel-gpu-plugin-4nfj2                                       Container created
+kube-system              4m55s       Normal    Created                          pod/local-path-provisioner-6858d854cf-fjp42                      Container created
+node-feature-discovery   4m55s       Normal    Started                          pod/nfd-master-74b44b767d-b4lcv                                  Container started
+kube-system              4m55s       Normal    Pulling                          pod/coredns-577d995dff-m4bm5                                     Pulling image "rancher/mirrored-coredns-coredns:1.14.7"
+node-feature-discovery   4m55s       Normal    Started                          pod/nfd-worker-2fbwn                                             Container started
+node-feature-discovery   4m55s       Normal    Created                          pod/nfd-worker-2fbwn                                             Container created
+node-feature-discovery   4m55s       Normal    Pulled                           pod/nfd-worker-2fbwn                                             Successfully pulled image "registry.k8s.io/nfd/node-feature-discovery:v0.20.1@sha256:6fc5c9af9176e1acb5ce57a4a1d0c0d3eb25cd34f5bdd0efefebf0efe45a8379" in 3.081s (3.081s including waiting). Image size: 87537491 bytes.
+node-feature-discovery   4m55s       Warning   Unhealthy                        pod/nfd-master-74b44b767d-b4lcv                                  Readiness probe failed: Get "http://10.42.0.3:8080/healthz": dial tcp 10.42.0.3:8080: connect: connection refused
+node-feature-discovery   4m55s       Normal    Created                          pod/nfd-master-74b44b767d-b4lcv                                  Container created
+kube-system              4m55s       Normal    Started                          pod/local-path-provisioner-6858d854cf-fjp42                      Container started
+kube-system              4m55s       Normal    Pulled                           pod/local-path-provisioner-6858d854cf-fjp42                      Successfully pulled image "rancher/local-path-provisioner:v0.0.37" in 2.496s (2.496s including waiting). Image size: 15066012 bytes.
+node-feature-discovery   4m55s       Normal    Pulled                           pod/nfd-master-74b44b767d-b4lcv                                  Successfully pulled image "registry.k8s.io/nfd/node-feature-discovery:v0.20.1@sha256:6fc5c9af9176e1acb5ce57a4a1d0c0d3eb25cd34f5bdd0efefebf0efe45a8379" in 3.08s (3.08s including waiting). Image size: 87537491 bytes.
+node-feature-discovery   4m55s       Normal    Started                          pod/nfd-gc-8d4595cbc-82fbz                                       Container started
+node-feature-discovery   4m55s       Normal    Created                          pod/nfd-gc-8d4595cbc-82fbz                                       Container created
+node-feature-discovery   4m55s       Normal    Pulled                           pod/nfd-gc-8d4595cbc-82fbz                                       Successfully pulled image "registry.k8s.io/nfd/node-feature-discovery:v0.20.1@sha256:6fc5c9af9176e1acb5ce57a4a1d0c0d3eb25cd34f5bdd0efefebf0efe45a8379" in 3.084s (3.084s including waiting). Image size: 87537491 bytes.
+kube-system              4m54s       Warning   Unhealthy                        pod/metrics-server-778c4649b4-rl6zq                              Readiness probe failed: Get "https://10.42.0.7:10250/readyz": dial tcp 10.42.0.7:10250: connect: connection refused
+kube-system              4m54s       Normal    Started                          pod/metrics-server-778c4649b4-rl6zq                              Container started
+kube-system              4m54s       Normal    Created                          pod/metrics-server-778c4649b4-rl6zq                              Container created
+kube-system              4m54s       Normal    Pulled                           pod/metrics-server-778c4649b4-rl6zq                              Successfully pulled image "rancher/mirrored-metrics-server:v0.9.0" in 2.247s (2.247s including waiting). Image size: 23258428 bytes.
+kube-system              4m53s       Normal    Pulled                           pod/helm-install-gateway-api-crd-8zw5k                           Successfully pulled image "rancher/klipper-helm:v0.13.3-build20260727" in 3.903s (3.903s including waiting). Image size: 74172006 bytes.
+kube-system              4m53s       Normal    Created                          pod/helm-install-gateway-api-crd-8zw5k                           Container created
+kube-system              4m52s       Normal    Pulled                           pod/coredns-577d995dff-m4bm5                                     Successfully pulled image "rancher/mirrored-coredns-coredns:1.14.7" in 3.097s (3.097s including waiting). Image size: 24261571 bytes.
+kube-system              4m52s       Normal    Created                          pod/coredns-577d995dff-m4bm5                                     Container created
+kube-system              4m52s       Normal    Started                          pod/coredns-577d995dff-m4bm5                                     Container started
+kube-system              4m52s       Warning   Unhealthy                        pod/coredns-577d995dff-m4bm5                                     Readiness probe failed: Get "http://10.42.0.9:8181/ready": dial tcp 10.42.0.9:8181: connect: connection refused
+kube-system              4m52s       Normal    Started                          pod/helm-install-gateway-api-crd-8zw5k                           Container started
+kube-system              4m48s       Normal    Completed                        job/helm-install-gateway-api-crd                                 Job completed
+k3d-gpu-agent            4m44s       Normal    SuccessfulCreate                 replicaset/vllm-78c46bdf8f                                       Created pod: vllm-78c46bdf8f-668sw
+k3d-gpu-agent            4m44s       Normal    WaitForFirstConsumer             persistentvolumeclaim/vllm-models                                waiting for first consumer to be created before binding
+kube-system              4m44s       Normal    Pulling                          pod/helper-pod-create-pvc-e99c16ed-b576-42e6-b2a6-400efd81f8d7   Pulling image "rancher/mirrored-library-busybox:1.37.0"
+k3d-gpu-agent            4m44s       Normal    Provisioning                     persistentvolumeclaim/vllm-models                                External provisioner is provisioning volume for claim "k3d-gpu-agent/vllm-models"
+k3d-gpu-agent            4m44s       Normal    ScalingReplicaSet                deployment/vllm                                                  Scaled up replica set vllm-78c46bdf8f from 0 to 1
+kube-system              4m42s       Normal    Pulled                           pod/helper-pod-create-pvc-e99c16ed-b576-42e6-b2a6-400efd81f8d7   Successfully pulled image "rancher/mirrored-library-busybox:1.37.0" in 1.491s (1.491s including waiting). Image size: 2151353 bytes.
+kube-system              4m42s       Normal    Created                          pod/helper-pod-create-pvc-e99c16ed-b576-42e6-b2a6-400efd81f8d7   Container created
+kube-system              4m42s       Normal    Started                          pod/helper-pod-create-pvc-e99c16ed-b576-42e6-b2a6-400efd81f8d7   Container started
+k3d-gpu-agent            4m42s       Normal    ExternalProvisioning             persistentvolumeclaim/vllm-models                                Waiting for a volume to be created either by the external provisioner 'rancher.io/local-path' or manually by the system administrator. If volume creation is delayed, please verify that the provisioner is running and correctly registered.
+kube-system              4m41s       Warning   Unhealthy                        pod/metrics-server-778c4649b4-rl6zq                              Readiness probe failed: HTTP probe failed with statuscode: 500
+k3d-gpu-agent            4m40s       Normal    ProvisioningSucceeded            persistentvolumeclaim/vllm-models                                Successfully provisioned volume pvc-e99c16ed-b576-42e6-b2a6-400efd81f8d7
+k3d-gpu-agent            4m39s       Normal    Pulling                          pod/vllm-78c46bdf8f-668sw                                        Pulling image "vllm/vllm-openai-xpu:v0.33.0@sha256:be3a9aea234be64ffc2816ac561c597e0f0b1f44a61a7f4912780e4d77450086"
+k3d-gpu-agent            2m25s       Normal    Pulled                           pod/vllm-78c46bdf8f-668sw                                        Successfully pulled image "vllm/vllm-openai-xpu:v0.33.0@sha256:be3a9aea234be64ffc2816ac561c597e0f0b1f44a61a7f4912780e4d77450086" in 2m13.447s (2m13.447s including waiting). Image size: 5212361514 bytes.
+k3d-gpu-agent            2m25s       Normal    Started                          pod/vllm-78c46bdf8f-668sw                                        Container started
+k3d-gpu-agent            2m25s       Normal    Created                          pod/vllm-78c46bdf8f-668sw                                        Container created
+k3d-gpu-agent            34s         Warning   Unhealthy                        pod/vllm-78c46bdf8f-668sw                                        Readiness probe failed: Get "http://10.42.0.11:8000/health": dial tcp 10.42.0.11:8000: connect: connection refused
diff --git a/bench/vllm-agent/results/2026-11-08/pr357-0985b179/k3d-containers-after.txt b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/k3d-containers-after.txt
new file mode 100644
index 00000000..e69de29b
diff --git a/bench/vllm-agent/results/2026-11-08/pr357-0985b179/models.json b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/models.json
new file mode 100644
index 00000000..6388c89a
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/models.json
@@ -0,0 +1 @@
+{"object":"list","data":[{"id":"Qwen3-4B","object":"model","created":1794119290,"owned_by":"vllm","root":"Qwen/Qwen3-4B","parent":null,"max_model_len":8192,"permission":[{"id":"modelperm-b93e759333731320","object":"model_permission","created":1794119290,"allow_create_engine":false,"allow_sampling":true,"allow_logprobs":true,"allow_search_indices":false,"allow_view":true,"allow_fine_tuning":false,"organization":"*","group":null,"is_blocking":false}]}]}
\ No newline at end of file
diff --git a/bench/vllm-agent/results/2026-11-08/pr357-0985b179/pf.log b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/pf.log
new file mode 100644
index 00000000..44076ae3
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/pf.log
@@ -0,0 +1,5 @@
+Forwarding from 127.0.0.1:8000 -> 8000
+Forwarding from [::1]:8000 -> 8000
+Handling connection for 8000
+Handling connection for 8000
+Handling connection for 8000
diff --git a/bench/vllm-agent/results/2026-11-08/pr357-0985b179/pods.txt b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/pods.txt
new file mode 100644
index 00000000..3b4e7c17
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/pods.txt
@@ -0,0 +1,8 @@
+NAME                   READY   UP-TO-DATE   AVAILABLE   AGE     CONTAINERS   IMAGES                                                                                                 SELECTOR
+deployment.apps/vllm   1/1     1            1           4m17s   vllm         vllm/vllm-openai-xpu:v0.33.0@sha256:be3a9aea234be64ffc2816ac561c597e0f0b1f44a61a7f4912780e4d77450086   app=vllm
+
+NAME                        READY   STATUS    RESTARTS   AGE     IP           NODE               NOMINATED NODE   READINESS GATES
+pod/vllm-78c46bdf8f-668sw   1/1     Running   0          4m17s   10.42.0.11   k3d-gpu-server-0   <none>           <none>
+
+NAME           TYPE        CLUSTER-IP     EXTERNAL-IP   PORT(S)    AGE     SELECTOR
+service/vllm   ClusterIP   10.43.99.150   <none>        8000/TCP   4m17s   app=vllm
diff --git a/bench/vllm-agent/results/2026-11-08/pr357-0985b179/tool-req.json b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/tool-req.json
new file mode 100644
index 00000000..af9846a3
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/tool-req.json
@@ -0,0 +1,33 @@
+{
+  "model": "Qwen3-4B",
+  "messages": [
+    {
+      "role": "user",
+      "content": "What is the weather in Paris right now?"
+    }
+  ],
+  "tools": [
+    {
+      "type": "function",
+      "function": {
+        "name": "get_weather",
+        "description": "Get the current weather for a city",
+        "parameters": {
+          "type": "object",
+          "properties": {
+            "city": {
+              "type": "string",
+              "description": "City name"
+            }
+          },
+          "required": [
+            "city"
+          ]
+        }
+      }
+    }
+  ],
+  "tool_choice": "auto",
+  "temperature": 0,
+  "max_tokens": 2048
+}
diff --git a/bench/vllm-agent/results/2026-11-08/pr357-0985b179/tool.json b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/tool.json
new file mode 100644
index 00000000..494c175b
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/tool.json
@@ -0,0 +1 @@
+{"id":"chatcmpl-a46abb18a1c46a54","object":"chat.completion","created":1794119313,"model":"Qwen3-4B","choices":[{"index":0,"message":{"role":"assistant","content":null,"refusal":null,"annotations":null,"audio":null,"function_call":null,"tool_calls":[{"id":"chatcmpl-tool-8874486735506104","type":"function","function":{"name":"get_weather","arguments":"{\"city\": \"Paris\"}"}}],"reasoning":"\nOkay, the user is asking for the current weather in Paris. I need to use the get_weather function. Let me check the function parameters. The function requires the city name as a string. The user mentioned \"Paris,\" so I'll set the city parameter to \"Paris\". I should make sure there are no typos. Alright, the tool call should be correct.\n"},"logprobs":null,"finish_reason":"tool_calls","stop_reason":null,"token_ids":null,"routed_experts":null,"weight_versions":null}],"service_tier":null,"system_fingerprint":"vllm-0.33.0-5f414ee0","usage":{"prompt_tokens":164,"total_tokens":263,"completion_tokens":99,"prompt_tokens_details":null,"completion_tokens_details":{"reasoning_tokens":76}},"prompt_logprobs":null,"prompt_token_ids":null,"prompt_text":null,"kv_transfer_params":null,"ec_transfer_params":null,"metrics":null}
\ No newline at end of file
diff --git a/bench/vllm-agent/results/2026-11-08/pr357-0985b179/tools.txt b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/tools.txt
new file mode 100644
index 00000000..ce715234
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/tools.txt
@@ -0,0 +1,6 @@
+/home/devcloud/.local/bin/k3d
+/home/devcloud/.local/bin/kubectl
+k3d version v5.9.0
+k3s version v1.35.5-k3s1 (default)
+Client Version: v1.37.2
+Kustomize Version: v5.8.1
diff --git a/bench/vllm-agent/results/2026-11-08/pr357-0985b179/up.txt b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/up.txt
new file mode 100644
index 00000000..adf5b674
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/up.txt
@@ -0,0 +1,97 @@
+WARN: present but disabled: DSA dsa0 dsa2 dsa4, QAT 0000:01:00.0 0000:06:00.0 0000:0b:00.0 (no user work queues / VFs), so not in the cluster — enable them: k3d-gpu host-setup --apply, then k3d-gpu down && k3d-gpu up (exact commands: k3d-gpu doctor)
+==> Accelerators: intel amx
+==> Experimental features (auto): dra image-volume
+==> Pulling node image cryptoandcoffee/k3d-gpu:0.1.95...
+0.1.95: Pulling from cryptoandcoffee/k3d-gpu
+3fc420076e28: Pulling fs layer
+c7246eb68d8e: Pulling fs layer
+f772acb95844: Pulling fs layer
+4c0d817598b2: Pulling fs layer
+e4f81e59a4dd: Pulling fs layer
+72fa97de23e4: Pulling fs layer
+5ce602c69cb9: Pulling fs layer
+40eccae86b19: Pulling fs layer
+951c4ec85de1: Pulling fs layer
+c6ceffcc1e6e: Pulling fs layer
+a2b45b3f52ac: Pulling fs layer
+1ea8a99cd78e: Pulling fs layer
+03b22aa06829: Pulling fs layer
+79a61545e121: Pulling fs layer
+098d2e9a74b9: Pulling fs layer
+df5798a4e788: Pulling fs layer
+c6ceffcc1e6e: Already exists
+951c4ec85de1: Already exists
+44136fa355b3: Already exists
+e4f81e59a4dd: Download complete
+57615e2ad07d: Download complete
+f772acb95844: Download complete
+c7246eb68d8e: Download complete
+3fc420076e28: Download complete
+df5798a4e788: Download complete
+5ce602c69cb9: Download complete
+a2b45b3f52ac: Download complete
+72fa97de23e4: Download complete
+1ea8a99cd78e: Download complete
+4c0d817598b2: Download complete
+03b22aa06829: Download complete
+79a61545e121: Download complete
+098d2e9a74b9: Download complete
+40eccae86b19: Download complete
+40eccae86b19: Pull complete
+c6ceffcc1e6e: Pull complete
+951c4ec85de1: Pull complete
+f772acb95844: Pull complete
+c7246eb68d8e: Pull complete
+5ce602c69cb9: Pull complete
+a2b45b3f52ac: Pull complete
+72fa97de23e4: Pull complete
+1ea8a99cd78e: Pull complete
+4c0d817598b2: Pull complete
+03b22aa06829: Pull complete
+c6ceffcc1e6e: Pull complete
+951c4ec85de1: Pull complete
+e4f81e59a4dd: Pull complete
+f772acb95844: Pull complete
+c7246eb68d8e: Pull complete
+3fc420076e28: Pull complete
+df5798a4e788: Pull complete
+5ce602c69cb9: Pull complete
+a2b45b3f52ac: Pull complete
+72fa97de23e4: Pull complete
+1ea8a99cd78e: Pull complete
+4c0d817598b2: Pull complete
+03b22aa06829: Pull complete
+79a61545e121: Pull complete
+098d2e9a74b9: Pull complete
+Digest: sha256:65499cbe202dca63c28029e03c4eb5139f5de31f5f369d915494f06e6f4dd3fa
+Status: Downloaded newer image for cryptoandcoffee/k3d-gpu:0.1.95
+docker.io/cryptoandcoffee/k3d-gpu:0.1.95
+==> Creating k3d cluster 'gpu' with image cryptoandcoffee/k3d-gpu:0.1.95...
+INFO[0000] Prep: Network                                
+INFO[0000] Created network 'k3d-gpu'                    
+INFO[0000] Created image volume k3d-gpu-images          
+INFO[0000] Starting new tools node...                   
+INFO[0000] Starting node 'k3d-gpu-tools'                
+INFO[0001] Creating node 'k3d-gpu-server-0'             
+INFO[0001] Using the k3d-tools node to gather environment information 
+INFO[0001] HostIP: using network gateway 172.19.0.1 address 
+INFO[0001] Starting cluster 'gpu'                       
+INFO[0001] Starting servers...                          
+INFO[0001] Starting node 'k3d-gpu-server-0'             
+INFO[0004] All agents already running.                  
+INFO[0004] All helpers already running.                 
+INFO[0004] Injecting records for hostAliases (incl. host.k3d.internal) and for 1 network members into CoreDNS configmap... 
+INFO[0006] Cluster 'gpu' created successfully!          
+INFO[0006] --kubeconfig-update-default=false --> sets --kubeconfig-switch-context=false 
+INFO[0006] You can now use it like this:                
+export KUBECONFIG=$(k3d kubeconfig write gpu)
+kubectl cluster-info
+==> kubectl for this cluster: export KUBECONFIG=/home/devcloud/.local/state/k3d-gpu/kubeconfig-gpu
+==> Waiting for nodes Ready...
+node/k3d-gpu-server-0 condition met
+==> Waiting for cluster DNS (CoreDNS)...
+==> intel device plugin auto-deployed by k3s.
+==> Verifying accelerators are schedulable...
+daemon set "intel-gpu-plugin" successfully rolled out
+WARN: present but disabled: DSA dsa0 dsa2 dsa4, QAT 0000:01:00.0 0000:06:00.0 0000:0b:00.0 (no user work queues / VFs), so not in the cluster — enable them: k3d-gpu host-setup --apply, then k3d-gpu down && k3d-gpu up (exact commands: k3d-gpu doctor)
+==> Accelerators schedulable: intel=4 amx=1. Try: k3d-gpu test
diff --git a/bench/vllm-agent/results/2026-11-08/pr357-0985b179/vllm-previous.log b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/vllm-previous.log
new file mode 100644
index 00000000..3f2b2aaa
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/vllm-previous.log
@@ -0,0 +1 @@
+Error from server (BadRequest): previous terminated container "vllm" in pod "vllm-78c46bdf8f-668sw" not found
diff --git a/bench/vllm-agent/results/2026-11-08/pr357-0985b179/vllm.log b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/vllm.log
new file mode 100644
index 00000000..964295a6
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/pr357-0985b179/vllm.log
@@ -0,0 +1,335 @@
+W1108 06:26:12.796000 1 torch/_opaque_base.py:6] torch._opaque_base is deprecated, use torch._custom_class_base instead
+W1108 06:26:12.797000 1 torch/_library/opaque_object.py:288] register_opaque_type is deprecated, use register_custom_class instead
+W1108 06:26:12.797000 1 torch/_library/opaque_object.py:211] typ='value' is deprecated, use typ='constant' instead
+(APIServer pid=1) INFO 11-08 06:26:17 [startup.py:38] 
+(APIServer pid=1) INFO 11-08 06:26:17 [startup.py:38]        █     █     █▄   ▄█
+(APIServer pid=1) INFO 11-08 06:26:17 [startup.py:38]  ▄▄ ▄█ █     █     █ ▀▄▀ █  version 0.33.0
+(APIServer pid=1) INFO 11-08 06:26:17 [startup.py:38]   █▄█▀ █     █     █     █  model   Qwen/Qwen3-4B
+(APIServer pid=1) INFO 11-08 06:26:17 [startup.py:38]    ▀▀  ▀▀▀▀▀ ▀▀▀▀▀ ▀     ▀
+(APIServer pid=1) INFO 11-08 06:26:17 [startup.py:38] 
+(APIServer pid=1) INFO 11-08 06:26:17 [startup.py:96] non-default args: {'model_tag': 'Qwen/Qwen3-4B', 'enable_auto_tool_choice': True, 'tool_call_parser': 'hermes', 'model': 'Qwen/Qwen3-4B', 'dtype': 'float16', 'revision': '1cfa9a7208912126459214e8b04321603b3df60c', 'max_model_len': 8192, 'enforce_eager': True, 'served_model_name': ['Qwen3-4B'], 'reasoning_parser': 'qwen3', 'block_size': 64, 'gpu_memory_utilization': 0.9, 'enable_prefix_caching': False, 'max_num_batched_tokens': 2048}
+Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.
+(APIServer pid=1) INFO 11-08 06:26:26 [model.py:744] Resolved architecture: Qwen3ForCausalLM
+(APIServer pid=1) WARNING 11-08 06:26:26 [model.py:2516] Casting torch.bfloat16 to torch.float16.
+(APIServer pid=1) INFO 11-08 06:26:26 [model.py:2170] Using max model len 8192
+
Parse safetensors files:   0%|          | 0/3 [00:00<?, ?it/s]
Parse safetensors files:  33%|███▎      | 1/3 [00:00<00:00,  7.14it/s]
Parse safetensors files: 100%|██████████| 3/3 [00:00<00:00, 19.22it/s]
+(APIServer pid=1) WARNING 11-08 06:26:27 [vllm.py:1778] Enforce eager set, disabling torch.compile, CUDAGraphs, and JIT kernel warmup. This is equivalent to setting -cc.mode=none -cc.cudagraph_mode=none and --kernel_config.enable_jit_warmup=False
+(APIServer pid=1) WARNING 11-08 06:26:27 [vllm.py:1802] Inductor compilation was disabled by user settings, optimizations settings that are only active during inductor compilation will be ignored.
+(APIServer pid=1) INFO 11-08 06:26:27 [kernel.py:480] Final IR op priority after setting platform defaults: IrOpPriorityConfig(rms_norm=['vllm_c', 'native'], fused_add_rms_norm=['vllm_c', 'native'], gelu_and_mul_sparse=['native'])
+(APIServer pid=1) INFO 11-08 06:26:27 [vllm.py:2005] Cudagraph is disabled under eager mode
+(APIServer pid=1) INFO 11-08 06:26:27 [xpu.py:470] XPU platform: set server shutdown_timeout=5.
+
Fetching 10 files:   0%|          | 0/10 [00:00<?, ?it/s]
Fetching 10 files:  40%|████      | 4/10 [00:00<00:00, 39.35it/s]
Fetching 10 files: 100%|██████████| 10/10 [00:00<00:00, 29.38it/s]
Fetching 10 files: 100%|██████████| 10/10 [00:00<00:00, 30.17it/s]
+(APIServer pid=1) INFO 11-08 06:26:28 [compilation.py:317] Enabled custom fusions: norm_quant, act_quant
+W1108 06:26:32.371000 578 torch/_opaque_base.py:6] torch._opaque_base is deprecated, use torch._custom_class_base instead
+W1108 06:26:32.371000 578 torch/_library/opaque_object.py:288] register_opaque_type is deprecated, use register_custom_class instead
+W1108 06:26:32.371000 578 torch/_library/opaque_object.py:211] typ='value' is deprecated, use typ='constant' instead
+(EngineCore pid=578) INFO 11-08 06:26:36 [core.py:139] Initializing a V1 LLM engine (v0.33.0) with config: model='Qwen/Qwen3-4B', speculative_config=None, tokenizer='Qwen/Qwen3-4B', skip_tokenizer_init=False, tokenizer_mode=auto, revision=1cfa9a7208912126459214e8b04321603b3df60c, tokenizer_revision=1cfa9a7208912126459214e8b04321603b3df60c, trust_remote_code=False, dtype=torch.float16, max_seq_len=8192, download_dir=None, load_format=auto, tensor_parallel_size=1, pipeline_parallel_size=1, data_parallel_size=1, decode_context_parallel_size=1, dcp_comm_backend=ag_rs, disable_custom_all_reduce=True, quantization=None, quantization_config=None, enforce_eager=True, aux_output_config=AuxOutputConfig(enable_return_routed_experts=False, max_bytes=None), kv_cache_dtype=auto, device_config=xpu, structured_outputs_config=StructuredOutputsConfig(backend='auto', disable_any_whitespace=False, disable_additional_properties=False, reasoning_parser='qwen3', reasoning_parser_plugin='', enable_in_reasoning=False), observability_config=ObservabilityConfig(show_hidden_metrics_for_version=None, otlp_traces_endpoint=None, collect_detailed_traces=None, per_request_spec_decode_metrics='none', kv_cache_metrics=False, kv_cache_metrics_sample=0.01, custom_histogram_buckets=None, cudagraph_metrics=False, enable_layerwise_nvtx_tracing=False, enable_mfu_metrics=False, enable_mm_processor_stats=False, enable_logging_iteration_details=False, jit_monitor_mode='warn', jit_monitor_verbose=False), seed=0, served_model_name=Qwen3-4B, enable_prefix_caching=False, enable_chunked_prefill=True, pooler_config=None, compilation_config={'mode': <CompilationMode.NONE: 0>, 'debug_dump_path': None, 'cache_dir': '', 'compile_cache_save_format': 'binary', 'backend': 'inductor', 'custom_ops': ['all'], 'ir_enable_torch_wrap': False, 'splitting_ops': [], 'compile_mm_encoder': False, 'cudagraph_mm_encoder': False, 'encoder_cudagraph_token_budgets': [], 'encoder_cudagraph_max_vision_items_per_batch': 0, 'encoder_cudagraph_max_frames_per_batch': None, 'cudagraph_decoder_replay': True, 'decoder_replay_cudagraph_capture_sizes': [], 'decoder_replay_trim_threshold': 768, 'compile_sizes': [], 'compile_ranges_endpoints': [2048], 'inductor_compile_config': {'enable_auto_functionalized_v2': False, 'combo_kernels': True, 'benchmark_combo_kernel': True}, 'inductor_passes': {}, 'cudagraph_mode': <CUDAGraphMode.NONE: 0>, 'cudagraph_num_of_warmups': 0, 'cudagraph_capture_sizes': [], 'cudagraph_copy_inputs': False, 'cudagraph_specialize_lora': True, 'use_inductor_graph_partition': False, 'pass_config': {'fuse_norm_quant': True, 'fuse_act_quant': True, 'fuse_attn_quant': False, 'fuse_allreduce_rms': False, 'enable_qk_norm_rope_fusion': False, 'fuse_rope_kvcache_cat_mla': False, 'fuse_act_padding': False, 'fuse_qk_norm_rope_kvcache': False}, 'max_cudagraph_capture_size': 0, 'dynamic_shapes_config': {'type': <DynamicShapesType.BACKED: 'backed'>, 'evaluate_guards': False, 'assume_32_bit_indexing': False}, 'local_cache_dir': None, 'fast_moe_cold_start': False, 'static_all_moe_layers': []}, kernel_config=KernelConfig(ir_op_priority=IrOpPriorityConfig(rms_norm=['vllm_c', 'native'], fused_add_rms_norm=['vllm_c', 'native'], gelu_and_mul_sparse=['native']), enable_flashinfer_autotune=True, enable_rocm_segmented_attn_autotune=False, enable_cutedsl_warmup=True, enable_jit_warmup=False, monokernel=False, moe_backend='auto', sparse_indexer_topk_backend='auto', linear_backend='auto', linear_backend_per_quant=None)
+(EngineCore pid=578) INFO 11-08 06:26:37 [parallel_state.py:1929] world_size=1 rank=0 local_rank=0 distributed_init_method=file:///tmp/vllm_dist_e78d05d2c0dd48e09dc4a0e9a8f752dc backend=xccl
+(EngineCore pid=578) INFO 11-08 06:26:37 [parallel_state.py:2385] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, PCP rank 0, TP rank 0, ETP rank 0, EP rank N/A, EPLB rank N/A
+(EngineCore pid=578) INFO 11-08 06:26:37 [xpu_worker.py:198] Using V2 Model Runner
+(EngineCore pid=578) INFO 11-08 06:26:37 [model_runner.py:399] Loading model from scratch...
+(EngineCore pid=578) INFO 11-08 06:26:38 [xpu.py:270] Using Flash Attention backend.
+(EngineCore pid=578) INFO 11-08 06:26:38 [flash_attn.py:1122] Using FlashAttention version 2
+(EngineCore pid=578) INFO 11-08 06:26:38 [base_loader.py:82] Initializing model took 0.949975 seconds
+Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.
+(EngineCore pid=578) INFO 11-08 06:27:49 [weight_utils.py:537] Time spent downloading weights for Qwen/Qwen3-4B: 70.435239 seconds
+(EngineCore pid=578) INFO 11-08 06:27:49 [weight_utils.py:1144] Filesystem type for checkpoints: EXT4. Checkpoint size: 7.49 GiB. Available RAM: 481.53 GiB.
+(EngineCore pid=578) INFO 11-08 06:27:49 [weight_utils.py:1167] Auto-prefetch is disabled because the filesystem (EXT4) is not a recognized network FS (NFS/Lustre). If you want to force prefetching, start vLLM with --safetensors-load-strategy=prefetch.
+
Loading safetensors checkpoint shards:   0% Completed | 0/3 [00:00<?, ?it/s]
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight embed_tokens.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.0.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.0.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.0.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.0.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.0.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.0.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.1.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.1.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.1.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.1.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.1.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.1.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.10.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.10.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.10.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.10.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.10.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.10.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.11.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.11.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.11.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.11.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.11.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.11.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.12.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.12.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.12.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.12.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.12.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.12.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.13.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.13.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.13.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.13.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.13.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.13.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.14.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.14.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.14.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.14.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.14.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.14.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.15.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.15.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.15.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.2.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.2.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.2.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.2.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.2.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.2.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.3.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.3.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.3.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.3.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.3.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.3.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.4.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.4.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.4.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.4.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.4.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.4.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.5.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:49 [utils.py:333] Attempted to load weight layers.5.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.5.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.5.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.5.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.5.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.6.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.6.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.6.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.6.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.6.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.6.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.7.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.7.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.7.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.7.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.7.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.7.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.8.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.8.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.8.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.8.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.8.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.8.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.9.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.9.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.9.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.9.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.9.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.9.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+
Loading safetensors checkpoint shards:  33% Completed | 1/3 [00:00<00:01,  1.56it/s]
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.15.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.15.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.15.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.16.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.16.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.16.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.16.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.16.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.16.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.17.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.17.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.17.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.17.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.17.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.17.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.18.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.18.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.18.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.18.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.18.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.18.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.19.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.19.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.19.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.19.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.19.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.19.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.20.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.20.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.20.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.20.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.20.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.20.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.21.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.21.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.21.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.21.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.21.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.21.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.22.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.22.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.22.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.22.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.22.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.22.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.23.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.23.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.23.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.23.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.23.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.23.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.24.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.24.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.24.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.24.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.24.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.24.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.25.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.25.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.25.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.25.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.25.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.25.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.26.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.26.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.26.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.26.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.26.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.26.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.27.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.27.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.27.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.27.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.27.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.27.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.28.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.28.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.28.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.28.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.28.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.28.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.29.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.29.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.29.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.29.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.29.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.29.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.30.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.30.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.30.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.30.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.30.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.30.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.31.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.31.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.31.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.31.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.31.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.31.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.32.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.32.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.32.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.32.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.32.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.32.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.33.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.33.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.33.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.33.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.33.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.33.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.34.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.34.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.34.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.34.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.34.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.34.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.35.self_attn.k_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.35.self_attn.o_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.35.self_attn.q_norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+
Loading safetensors checkpoint shards:  67% Completed | 2/3 [00:01<00:00,  1.59it/s]
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.35.input_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.35.mlp.down_proj.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight layers.35.post_attention_layernorm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+(EngineCore pid=578) WARNING 11-08 06:27:50 [utils.py:333] Attempted to load weight norm.weight with dtype torch.bfloat16 into parameter with dtype torch.float16
+
Loading safetensors checkpoint shards: 100% Completed | 3/3 [00:01<00:00,  2.35it/s]
+
+(EngineCore pid=578) INFO 11-08 06:27:50 [default_loader.py:506] Loading weights took 1.28 seconds
+(EngineCore pid=578) INFO 11-08 06:27:51 [model_runner.py:428] Model loading took 7.56 GiB memory and 73.904500 seconds
+(EngineCore pid=578) INFO 11-08 06:27:51 [topk_topp_sampler.py:182] Using the fused XPU kernel for top-p & top-k sampling.
+(EngineCore pid=578) INFO 11-08 06:27:51 [torch_utils.py:281] Reducing Torch threads from 48 to 1 for serving. Set OMP_NUM_THREADS in the external environment to override.
+(EngineCore pid=578) INFO 11-08 06:27:51 [utils.py:381] Using LBNHC KV cache layout.
+(EngineCore pid=578) INFO 11-08 06:28:01 [gpu_worker.py:772] Available KV cache memory: 18.1 GiB
+(EngineCore pid=578) INFO 11-08 06:28:01 [kv_cache_utils.py:1147] XPU KV cache size: 131,776 tokens, Maximum concurrency for 8,192 tokens per request: 16.09x
+(EngineCore pid=578) INFO 11-08 06:28:01 [gpu_worker.py:1066] Free memory on device (29.91/30.3 GiB) on startup. Desired GPU memory utilization is (0.9, 27.27 GiB). Actual usage is 8.79 GiB for consumed memory (weights + non-torch), 0.37 GiB for peak activation, and 0.0 GiB for CUDAGraph memory. Replace gpu_memory_utilization config with `--kv-cache-memory=19276367770` (17.95 GiB) to fit into requested memory, or `--kv-cache-memory=22112253440` (20.59 GiB) to fully utilize gpu memory. Current kv cache memory in use is 18.1 GiB.
+(EngineCore pid=578) INFO 11-08 06:28:02 [core.py:412] init engine (profile, create kv cache, warmup model) took 10.52 s
+
Fetching 10 files:   0%|          | 0/10 [00:00<?, ?it/s]
Fetching 10 files: 100%|██████████| 10/10 [00:00<00:00, 1744.86it/s]
+
Parse safetensors files:   0%|          | 0/3 [00:00<?, ?it/s]
Parse safetensors files:  33%|███▎      | 1/3 [00:00<00:00,  6.80it/s]
Parse safetensors files: 100%|██████████| 3/3 [00:00<00:00, 17.00it/s]
+(EngineCore pid=578) WARNING 11-08 06:28:02 [vllm.py:1778] Enforce eager set, disabling torch.compile, CUDAGraphs, and JIT kernel warmup. This is equivalent to setting -cc.mode=none -cc.cudagraph_mode=none and --kernel_config.enable_jit_warmup=False
+(EngineCore pid=578) WARNING 11-08 06:28:02 [vllm.py:1802] Inductor compilation was disabled by user settings, optimizations settings that are only active during inductor compilation will be ignored.
+(EngineCore pid=578) INFO 11-08 06:28:02 [kernel.py:480] Final IR op priority after setting platform defaults: IrOpPriorityConfig(rms_norm=['vllm_c', 'native'], fused_add_rms_norm=['vllm_c', 'native'], gelu_and_mul_sparse=['native'])
+(EngineCore pid=578) INFO 11-08 06:28:02 [vllm.py:2005] Cudagraph is disabled under eager mode
+(EngineCore pid=578) INFO 11-08 06:28:02 [compilation.py:317] Enabled custom fusions: norm_quant, act_quant
+(APIServer pid=1) INFO 11-08 06:28:02 [entry.py:135] Supported tasks: ['generate']
+(APIServer pid=1) INFO 11-08 06:28:02 [factories.py:105] Scale-out endpoints are disabled. Restart vllm serve with --enable-scale-out to enable them.
+(APIServer pid=1) INFO 11-08 06:28:03 [parser_manager.py:36] "auto" tool choice has been enabled.
+(APIServer pid=1) INFO 11-08 06:28:03 [hf.py:736] Detected the chat template content format to be 'string'. You can set `--chat-template-content-format` to override this.
+(APIServer pid=1) WARNING 11-08 06:28:03 [model.py:1907] Default vLLM sampling parameters have been overridden by the model's `generation_config.json`: `{'temperature': 0.6, 'top_k': 20, 'top_p': 0.95}`. If this is not intended, please relaunch vLLM instance with `--generation-config vllm`.
+(APIServer pid=1) INFO 11-08 06:28:03 [entry.py:139] Starting vLLM server on http://0.0.0.0:8000
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:188] Available routes are:
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /openapi.json, Methods: GET, HEAD
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /docs, Methods: GET, HEAD
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /docs/oauth2-redirect, Methods: GET, HEAD
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /redoc, Methods: GET, HEAD
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /load, Methods: GET
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /version, Methods: GET
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /kv_event_sources, Methods: GET
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /health, Methods: GET
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /metrics, Methods: GET
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /tokenize, Methods: POST
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /detokenize, Methods: POST
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /v1/models, Methods: GET
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /v1/systemone, Methods: POST
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /v1/decisions, Methods: POST
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /ping, Methods: GET
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /ping, Methods: POST
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /invocations, Methods: POST
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /v1/chat/completions, Methods: POST
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /v1/chat/completions/batch, Methods: POST
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /v1/responses, Methods: POST
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /v1/responses/{response_id}, Methods: GET
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /v1/responses/{response_id}/cancel, Methods: POST
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /v1/completions, Methods: POST
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /v1/messages, Methods: POST
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /v1/messages/count_tokens, Methods: POST
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /generative_scoring, Methods: POST
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /scale_elastic_ep, Methods: POST
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /is_scaling_elastic_ep, Methods: POST
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /v1/chat/completions/render, Methods: POST
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /v1/messages/render, Methods: POST
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /v1/completions/render, Methods: POST
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /v1/responses/render, Methods: POST
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /v1/chat/completions/derender, Methods: POST
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /v1/completions/derender, Methods: POST
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /inference/v1/generate, Methods: POST
+(APIServer pid=1) INFO 11-08 06:28:03 [launcher.py:197] Route: /inference/v1/abort_requests, Methods: POST
+INFO:     Started server process [1]
+INFO:     Waiting for application startup.
+INFO:     Application startup complete.
+INFO:     10.42.0.1:53044 - "GET /health HTTP/1.1" 200 OK
+INFO:     127.0.0.1:56376 - "POST /v1/chat/completions HTTP/1.1" 200 OK
+INFO:     127.0.0.1:56398 - "GET /v1/models HTTP/1.1" 200 OK
+INFO:     10.42.0.1:53056 - "GET /health HTTP/1.1" 200 OK
+(APIServer pid=1) INFO 11-08 06:28:13 [loggers.py:335] Engine 000: Avg prompt throughput: 3.6 tokens/s, Avg generation throughput: 16.0 tokens/s, Running: 1 reqs, Waiting: 0 reqs, XPU KV cache usage: 0.1%
+INFO:     10.42.0.1:59152 - "GET /health HTTP/1.1" 200 OK
+INFO:     10.42.0.1:59166 - "GET /health HTTP/1.1" 200 OK
+(APIServer pid=1) INFO 11-08 06:28:23 [loggers.py:335] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 63.8 tokens/s, Running: 1 reqs, Waiting: 0 reqs, XPU KV cache usage: 0.6%
+INFO:     10.42.0.1:54754 - "GET /health HTTP/1.1" 200 OK
+INFO:     10.42.0.1:54764 - "GET /health HTTP/1.1" 200 OK
+(APIServer pid=1) INFO 11-08 06:28:33 [loggers.py:335] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 61.2 tokens/s, Running: 1 reqs, Waiting: 0 reqs, XPU KV cache usage: 1.1%
+INFO:     127.0.0.1:56402 - "POST /v1/chat/completions HTTP/1.1" 200 OK
+INFO:     127.0.0.1:39232 - "POST /v1/chat/completions HTTP/1.1" 200 OK
diff --git a/bench/vllm-agent/results/2026-11-08/queue b/bench/vllm-agent/results/2026-11-08/queue
new file mode 100644
index 00000000..40846f74
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/queue
@@ -0,0 +1,2 @@
+main-0.1.94=/usr/bin/k3d-gpu
+pr357-0985b179=/home/devcloud/k3dgpu-bench/k3d-gpu-pr357-0985b179/scripts/k3d-gpu
diff --git a/bench/vllm-agent/results/2026-11-08/scripts/checkout.sh b/bench/vllm-agent/results/2026-11-08/scripts/checkout.sh
new file mode 100644
index 00000000..d41e9e53
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/scripts/checkout.sh
@@ -0,0 +1,13 @@
+#!/bin/bash
+# checkout.sh <sha> <dir>: a clean checkout of one commit of 88plug/k3d-gpu, unmodified.
+set -eu
+SHA=$1 D=$2
+rm -rf "$D"
+git init -q "$D"
+git -C "$D" fetch -q --depth 1 https://github.com/88plug/k3d-gpu.git "$SHA"
+git -C "$D" checkout -q FETCH_HEAD
+git -C "$D" log -1 --format='%H %s'
+git -C "$D" status --porcelain | wc -l
+grep -m1 '^VERSION=' "$D/scripts/k3d-gpu"
+grep -m1 'vllm/vllm-openai-xpu:v' "$D/scripts/k3d-gpu"
+sha256sum "$D/scripts/k3d-gpu"
diff --git a/bench/vllm-agent/results/2026-11-08/scripts/hold.sh b/bench/vllm-agent/results/2026-11-08/scripts/hold.sh
new file mode 100644
index 00000000..666aa5eb
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/scripts/hold.sh
@@ -0,0 +1,42 @@
+#!/bin/bash
+# hold.sh <run root> <label>=<launcher>...: under ~/k3dgpu-bench.lock (waits up to 12 h),
+# stop qwen38-a-hr, qwen38-b-hr and aiperf010 (takeover-stop.sh), run each launcher's
+# end-state check (run.sh), then the ones appended to <run root>/queue (one
+# <label>=<launcher> per line) until <run root>/release exists or the queue has been idle
+# for 2 h, then start what was stopped and wait for its baseline answers (restore.sh).
+# The lock is released when this script exits.
+set -u
+ROOT=$1
+shift
+BIN=$(cd "$(dirname "$0")" && pwd)
+mkdir -p "$ROOT" && cd "$ROOT" || exit 1
+log() { echo "$(date -u +%FT%TZ) $*" | tee -a "$ROOT/hold.log"; }
+echo $$ > hold.pid
+log "waiting for ~/k3dgpu-bench.lock"
+exec 9> "$HOME/k3dgpu-bench.lock"
+flock -w 43200 9 || { log "lock not taken in 12 h"; exit 3; }
+log "lock taken"
+bash "$BIN/takeover-stop.sh" "$ROOT/takeover" 9>&- >> "$ROOT/hold.log" 2>&1
+log "stopped: $(tr '\n' ' ' < "$ROOT/takeover/stopped.txt")"
+touch queue
+for spec in "$@"; do echo "${spec}" >> queue; done
+done_n=0 idle=${SECONDS}
+while :; do
+    n=$(wc -l < queue)
+    if [ "${n}" -gt "${done_n}" ]; then
+        spec=$(sed -n "$((done_n + 1))p" queue)
+        done_n=$((done_n + 1))
+        label=${spec%%=*} k=${spec#*=}
+        log "run ${label} with ${k}"
+        bash "$BIN/run.sh" "${label}" "${k}" "$ROOT/${label}" 9>&- >> "$ROOT/hold.log" 2>&1
+        log "run ${label} rc=$?"
+        idle=${SECONDS}
+        continue
+    fi
+    if [ -e release ]; then log "release asked"; break; fi
+    if [ $((SECONDS - idle)) -ge 7200 ]; then log "queue idle 2 h"; break; fi
+    sleep 20
+done
+bash "$BIN/restore.sh" "$ROOT/takeover" 9>&- >> "$ROOT/hold.log" 2>&1
+log "restore rc=$?"
+log "lock released"
diff --git a/bench/vllm-agent/results/2026-11-08/scripts/install-release.sh b/bench/vllm-agent/results/2026-11-08/scripts/install-release.sh
new file mode 100644
index 00000000..cffd6b33
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/scripts/install-release.sh
@@ -0,0 +1,14 @@
+#!/bin/bash
+# install-release.sh <dir>: the public installer from main, as the README gives it.
+set -u
+D=$1
+mkdir -p "$D" && cd "$D" || exit 1
+{ ls -la "$HOME/.local/bin"; dpkg -l k3d-gpu 2>&1; command -v k3d kubectl k3d-gpu; } > before.txt 2>&1
+curl -fsSLO https://raw.githubusercontent.com/88plug/k3d-gpu/main/packaging/install.sh || exit 1
+sha256sum install.sh > install.sh.sha256
+grep -m1 '^K3D_GPU_VERSION=' install.sh
+bash install.sh > install.log 2>&1
+echo "install rc=$?"
+tail -n 12 install.log
+{ ls -la "$HOME/.local/bin"; dpkg -l k3d-gpu 2>&1; } > after.txt 2>&1
+bash -lc 'command -v k3d-gpu k3d kubectl; k3d-gpu version; k3d version | head -1; kubectl version --client | head -1'
diff --git a/bench/vllm-agent/results/2026-11-08/scripts/restore.sh b/bench/vllm-agent/results/2026-11-08/scripts/restore.sh
new file mode 100644
index 00000000..3c2cc089
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/scripts/restore.sh
@@ -0,0 +1,44 @@
+#!/bin/bash
+# restore.sh <dir>: start what takeover-stop.sh stopped (qwen38-a-hr, qwen38-b-hr,
+# aiperf010, in that order), then wait (30 min ceiling each) until each server answers
+# its baseline request: content "Paris", finish_reason stop. Exit 0 only when all do.
+set -u
+T=$1
+cd "$T" || exit 1
+umask 077
+log() { echo "$(date -u +%FT%TZ) $*" | tee -a "$T/restore-log.txt"; }
+vram() { curl -s -m 10 localhost:9101/metrics | awk '/^xe_vram_used_bytes/ { match($0, /card="[^"]+"/); printf "%s %.2f;", substr($0, RSTART + 6, RLENGTH - 7), $2 / 2^30 }'; }
+REQ='{"model":"Qwen3.8-27B","messages":[{"role":"user","content":"What is the capital of France? Answer in one word."}],"max_tokens":48,"temperature":0,"chat_template_kwargs":{"enable_thinking":false}}'
+rc=0
+for n in qwen38-a-hr qwen38-b-hr aiperf010; do
+    grep -qx "$n" stopped.txt || continue
+    log "$n start rc=$(docker start "$n" > /dev/null 2>&1; echo $?)"
+done
+for n in qwen38-a-hr qwen38-b-hr aiperf010; do
+    grep -qx "$n" stopped.txt || continue
+    case "$n" in qwen38-a-hr) p=30000 ;; qwen38-b-hr) p=30001 ;; *) p=- ;; esac
+    if [ "$p" = - ]; then
+        sleep 5
+        log "$n $(docker inspect -f 'status={{.State.Status}} started={{.State.StartedAt}} restart={{.HostConfig.RestartPolicy.Name}}' "$n")"
+        [ "$(docker inspect -f '{{.State.Status}}' "$n")" = running ] || rc=1
+        continue
+    fi
+    t0=$(date +%s)
+    while :; do
+        el=$(( $(date +%s) - t0 ))
+        if [ "$(docker inspect -f '{{.State.Status}}' "$n")" != running ]; then log "FAIL $n is not running after ${el}s"; rc=1; break; fi
+        m=$(curl -s -m 120 -o "restore-$p.json" -w "http=%{http_code} total_s=%{time_total}" "localhost:$p/v1/chat/completions" -H 'Content-Type: application/json' -d "${REQ}")
+        c=$(jq -r '.choices[0].message.content // empty' "restore-$p.json" 2>/dev/null)
+        f=$(jq -r '.choices[0].finish_reason // empty' "restore-$p.json" 2>/dev/null)
+        if [ "$c" = Paris ] && [ "$f" = stop ]; then
+            log "OK $n port=$p ${m} $(jq -c '{model,content:.choices[0].message.content,finish:.choices[0].finish_reason,usage}' "restore-$p.json") after ${el}s"
+            break
+        fi
+        if [ "${el}" -ge 1800 ]; then log "FAIL $n port=$p no baseline answer in ${el}s (${m} content=${c:-none})"; rc=1; break; fi
+        sleep 15
+    done
+done
+sleep 5
+log "vram after restore: $(vram)"
+log "restore rc=${rc}"
+exit "${rc}"
diff --git a/bench/vllm-agent/results/2026-11-08/scripts/run.sh b/bench/vllm-agent/results/2026-11-08/scripts/run.sh
new file mode 100644
index 00000000..42dc63e4
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/scripts/run.sh
@@ -0,0 +1,97 @@
+#!/bin/bash
+# run.sh <label> <launcher> <out dir>: the unpatched end state of the vLLM agent on one
+# Arc Pro B70. 'k3d-gpu up', then 'K3D_GPU_AGENT_BACKEND=vllm k3d-gpu agent up' with the
+# default image and model, then two requests through the endpoint agent up prints (its
+# port-forward): "What is 17 times 23?" and one tool call. Tears the cluster down after.
+set -u
+L=$1 K=$2 O=$3
+mkdir -p "$O" && cd "$O" || exit 1
+export PATH="$HOME/.local/bin:$PATH"
+NS=k3d-gpu-agent
+log() { echo "$(date -u +%FT%TZ) [$L] $*" | tee -a "$O/driver.log"; }
+vram() { curl -s -m 10 localhost:9101/metrics | awk '/^xe_vram_used_bytes/ { match($0, /bdf="[^"]+"/); b = substr($0, RSTART + 5, RLENGTH - 6); match($0, /card="[^"]+"/); printf "%s %s %.2f;", b, substr($0, RSTART + 6, RLENGTH - 7), $2 / 2^30 }'; }
+strip() { sed 's/\x1b\[[0-9;]*m//g'; }
+PF=""
+teardown() {
+    [ -z "${PF}" ] || { kill "${PF}" 2>/dev/null; wait "${PF}" 2>/dev/null; }
+    if [ -n "${KC:-}" ] && [ -f "${KC}" ]; then
+        kubectl --kubeconfig "${KC}" -n "${NS}" logs deploy/vllm -c vllm > vllm.log 2>&1
+        kubectl --kubeconfig "${KC}" -n "${NS}" logs deploy/vllm -c vllm --previous > vllm-previous.log 2>&1
+        kubectl --kubeconfig "${KC}" get events -A --sort-by=.lastTimestamp > events.txt 2>&1
+    fi
+    "$K" agent down > agent-down.txt 2>&1; log "agent down rc=$?"
+    "$K" down > down.txt 2>&1; log "down rc=$?"
+    sleep 5
+    log "vram after: $(vram)"
+    docker ps -a --format '{{.Names}} {{.Status}}' | grep -i k3d > k3d-containers-after.txt
+    log "k3d containers after: $(wc -l < k3d-containers-after.txt)"
+}
+
+log "launcher $K: $("$K" version 2>&1)"
+command -v k3d kubectl > tools.txt; k3d version >> tools.txt 2>&1; kubectl version --client >> tools.txt 2>&1
+log "vram pre: $(vram)"
+t=${SECONDS}; "$K" up > up.txt 2>&1; rc=$?
+log "up rc=${rc} in $((SECONDS - t)) s"
+KC=$(strip < up.txt | sed -n 's/.*export KUBECONFIG=\([^ ]*\).*/\1/p' | tail -1)
+log "kubeconfig ${KC:-none}"
+if [ "${rc}" -ne 0 ] || [ -z "${KC}" ]; then teardown; log "RESULT FAIL (up)"; exit 1; fi
+kubectl --kubeconfig "${KC}" get nodes -o jsonpath='{.items[0].status.allocatable}' > allocatable.json
+log "allocatable $(cat allocatable.json)"
+
+t=${SECONDS}; K3D_GPU_AGENT_BACKEND=vllm "$K" agent up > agent-up.txt 2>&1; arc=$?
+log "agent up rc=${arc} in $((SECONDS - t)) s"
+strip < agent-up.txt > agent-up.clean.txt
+WARM=$(grep -a 'warm-up reply' agent-up.clean.txt | sed 's/^==> //')
+log "launcher warm-up line: ${WARM:-none}"
+kubectl --kubeconfig "${KC}" -n "${NS}" get deploy,pods,svc -o wide > pods.txt 2>&1
+kubectl --kubeconfig "${KC}" -n "${NS}" get deploy vllm -o yaml > deploy-vllm.yaml 2>&1
+kubectl --kubeconfig "${KC}" -n "${NS}" describe pods > describe-pods.txt 2>&1
+READY=$(kubectl --kubeconfig "${KC}" -n "${NS}" get pod -l app=vllm -o jsonpath='{.items[0].status.conditions[?(@.type=="Ready")].status}' 2>&1)
+LIM=$(kubectl --kubeconfig "${KC}" -n "${NS}" get pod -l app=vllm -o jsonpath='{.items[0].spec.containers[0].resources.limits}' 2>&1)
+IMG=$(kubectl --kubeconfig "${KC}" -n "${NS}" get pod -l app=vllm -o jsonpath='{.items[0].spec.containers[0].image} {.items[0].status.containerStatuses[0].imageID} restarts={.items[0].status.containerStatuses[0].restartCount}' 2>&1)
+log "pod Ready=${READY} limits=${LIM}"
+log "pod image ${IMG}"
+log "pod enableServiceLinks=$(kubectl --kubeconfig "${KC}" -n "${NS}" get deploy vllm -o jsonpath='{.spec.template.spec.enableServiceLinks}')"
+log "pod render nodes: $(kubectl --kubeconfig "${KC}" -n "${NS}" exec deploy/vllm -c vllm -- ls /dev/dri 2>&1 | tr '\n' ' ')"
+log "pod VLLM_PORT lines: $(kubectl --kubeconfig "${KC}" -n "${NS}" exec deploy/vllm -c vllm -- env 2>&1 | grep -c '^VLLM_PORT')"
+log "vram serving: $(vram)"
+if [ "${arc}" -ne 0 ]; then teardown; log "RESULT FAIL (agent up rc ${arc})"; exit 1; fi
+
+# The endpoint agent up prints: "<KUBECONFIG=...> kubectl -n k3d-gpu-agent port-forward svc/vllm 8000:8000".
+PFCMD=$(grep -a 'port-forward svc/vllm' agent-up.clean.txt | tail -1 | sed 's/^ *//')
+log "endpoint command: ${PFCMD}"
+# shellcheck disable=SC2086 # the printed command's words
+env ${PFCMD} > pf.log 2>&1 &
+PF=$!
+for _ in $(seq 1 30); do curl -s -m 5 localhost:8000/v1/models > models.json 2>/dev/null && break; sleep 2; done
+log "models: $(jq -c '[.data[].id]' models.json 2>&1)"
+
+MODEL=$(jq -r '.data[0].id' models.json)
+jq -n --arg m "${MODEL}" '{model:$m,messages:[{role:"user",content:"What is 17 times 23?"}],temperature:0,max_tokens:4096}' > chat-req.json
+m=$(curl -s -m 600 -o chat.json -w 'http=%{http_code} total_s=%{time_total}' localhost:8000/v1/chat/completions -H 'Content-Type: application/json' -d @chat-req.json)
+CONTENT=$(jq -r '.choices[0].message.content' chat.json 2>/dev/null)
+log "chat ${m} finish=$(jq -r '.choices[0].finish_reason' chat.json 2>/dev/null) usage=$(jq -c .usage chat.json 2>/dev/null)"
+log "chat content: $(tr '\n' ' ' <<< "${CONTENT}")"
+log "chat reasoning chars: $(jq -r '(.choices[0].message.reasoning_content // .choices[0].message.reasoning // "") | length' chat.json 2>/dev/null)"
+
+jq -n --arg m "${MODEL}" '{model:$m,messages:[{role:"user",content:"What is the weather in Paris right now?"}],
+  tools:[{type:"function",function:{name:"get_weather",description:"Get the current weather for a city",
+  parameters:{type:"object",properties:{city:{type:"string",description:"City name"}},required:["city"]}}}],
+  tool_choice:"auto",temperature:0,max_tokens:2048}' > tool-req.json
+m=$(curl -s -m 600 -o tool.json -w 'http=%{http_code} total_s=%{time_total}' localhost:8000/v1/chat/completions -H 'Content-Type: application/json' -d @tool-req.json)
+TOOL=$(jq -c '.choices[0].message.tool_calls' tool.json 2>/dev/null)
+ARGS=$(jq -r '.choices[0].message.tool_calls[0].function.arguments' tool.json 2>/dev/null)
+log "tool ${m} finish=$(jq -r '.choices[0].finish_reason' tool.json 2>/dev/null) tool_calls=${TOOL}"
+log "tool arguments parse: $(jq -c . <<< "${ARGS}" 2>&1)"
+
+pass=1
+[ "${READY}" = True ] || pass=0
+grep -q '"gpu.intel.com/xe":"1"' <<< "${LIM}" || pass=0
+[ -n "${WARM}" ] || pass=0
+grep -qw 391 <<< "${CONTENT}" || pass=0
+[ "$(jq -r '.choices[0].message.tool_calls[0].function.name' tool.json 2>/dev/null)" = get_weather ] || pass=0
+[ "$(jq -r '.city' <<< "${ARGS}" 2>/dev/null)" = Paris ] || pass=0
+teardown
+if [ "${pass}" = 1 ]; then log "RESULT PASS"; exit 0; fi
+log "RESULT FAIL (Ready=${READY}, warm-up=${WARM:+yes}, content has 391: $(grep -qw 391 <<< "${CONTENT}" && echo yes || echo no), tool=${TOOL})"
+exit 1
diff --git a/bench/vllm-agent/results/2026-11-08/scripts/takeover-stop.sh b/bench/vllm-agent/results/2026-11-08/scripts/takeover-stop.sh
new file mode 100644
index 00000000..4fd66714
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/scripts/takeover-stop.sh
@@ -0,0 +1,39 @@
+#!/bin/bash
+# takeover-stop.sh <dir>: snapshot qwen38-a-hr, qwen38-b-hr and aiperf010, answer each
+# server's baseline request, then stop the running ones gracefully (docker stop -t 120)
+# in the order of the 2026-11-07 takeover. Writes <dir>/stopped.txt (names stopped).
+set -u
+T=$1
+mkdir -p "$T" && cd "$T" || exit 1
+umask 077
+log() { echo "$(date -u +%FT%TZ) $*" | tee -a "$T/stop-log.txt"; }
+vram() { curl -s -m 10 localhost:9101/metrics | awk '/^xe_vram_used_bytes/ { match($0, /card="[^"]+"/); printf "%s %.2f;", substr($0, RSTART + 6, RLENGTH - 7), $2 / 2^30 }'; }
+REQ='{"model":"Qwen3.8-27B","messages":[{"role":"user","content":"What is the capital of France? Answer in one word."}],"max_tokens":48,"temperature":0,"chat_template_kwargs":{"enable_thinking":false}}'
+NAMES="aiperf010 qwen38-b-hr qwen38-a-hr"
+
+date -u +%FT%TZ > snapshot-time.txt
+docker ps -a --no-trunc > docker-ps-a.txt
+for n in ${NAMES}; do docker inspect "$n" > "inspect-$n.json" 2>&1; done
+curl -s -m 10 localhost:9101/metrics > xe-exporter-before.prom
+log "vram before: $(vram)"
+for n in qwen38-a-hr qwen38-b-hr; do
+    docker logs --since 1h "$n" > "logs-1h-$n.txt" 2>&1
+    log "$n POST lines in the last 10 min: $(docker logs --since 10m "$n" 2>&1 | grep -c 'POST /v1')"
+done
+for p in 30000 30001; do
+    m=$(curl -s -m 90 -o "baseline-$p.json" -w "port=$p http=%{http_code} total_s=%{time_total}" "localhost:$p/v1/chat/completions" -H 'Content-Type: application/json' -d "${REQ}")
+    log "baseline ${m} $(jq -c '{model,content:.choices[0].message.content,finish:.choices[0].finish_reason}' "baseline-$p.json" 2>/dev/null)"
+done
+: > stopped.txt
+for n in ${NAMES}; do
+    st=$(docker inspect -f '{{.State.Status}}' "$n" 2>/dev/null)
+    if [ "${st}" != running ]; then log "$n not running (${st:-absent}), left as is"; continue; fi
+    log "$n stop begin"
+    t=$(date +%s)
+    docker stop -t 120 "$n" > /dev/null
+    log "$n stop took $(( $(date +%s) - t ))s: $(docker inspect -f 'status={{.State.Status}} exit={{.State.ExitCode}} restart={{.HostConfig.RestartPolicy.Name}}' "$n")"
+    echo "$n" >> stopped.txt
+done
+sleep 10
+curl -s -m 10 localhost:9101/metrics > xe-exporter-after.prom
+log "vram after: $(vram)"
diff --git a/bench/vllm-agent/results/2026-11-08/scripts/uninstall-release.sh b/bench/vllm-agent/results/2026-11-08/scripts/uninstall-release.sh
new file mode 100644
index 00000000..1fe1b5a1
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/scripts/uninstall-release.sh
@@ -0,0 +1,13 @@
+#!/bin/bash
+# uninstall-release.sh <dir>: back to the state before install-release.sh: the k3d-gpu
+# package, and the k3d and kubectl install.sh put in ~/.local/bin (neither was there).
+set -u
+D=$1
+cd "$D" || exit 1
+# shellcheck disable=SC2024 # the log is the user's file: only apt-get runs as root
+sudo DEBIAN_FRONTEND=noninteractive NEEDRESTART_MODE=l apt-get remove -y k3d-gpu > uninstall.log 2>&1
+echo "apt-get remove rc=$?"
+rm -f "$HOME/.local/bin/k3d" "$HOME/.local/bin/kubectl"
+rmdir "$HOME/.local/state/k3d-gpu" 2>/dev/null
+{ ls -la "$HOME/.local/bin"; dpkg -l k3d-gpu 2>&1; command -v k3d-gpu k3d kubectl; } > after-uninstall.txt 2>&1
+cat after-uninstall.txt
diff --git a/bench/vllm-agent/results/2026-11-08/takeover-restore.txt b/bench/vllm-agent/results/2026-11-08/takeover-restore.txt
new file mode 100644
index 00000000..a00d836d
--- /dev/null
+++ b/bench/vllm-agent/results/2026-11-08/takeover-restore.txt
@@ -0,0 +1,23 @@
+2026-11-08T05:51:59Z vram before: card1 15.67;card2 14.97;card3 15.67;card4 14.97;
+2026-11-08T05:53:41Z qwen38-a-hr POST lines in the last 10 min: 0
+2026-11-08T05:55:21Z qwen38-b-hr POST lines in the last 10 min: 0
+2026-11-08T05:55:21Z baseline port=30000 http=200 total_s=0.154155 {"model":"Qwen3.8-27B","content":"Paris","finish":"stop"}
+2026-11-08T05:55:21Z baseline port=30001 http=200 total_s=0.152700 {"model":"Qwen3.8-27B","content":"Paris","finish":"stop"}
+2026-11-08T05:55:21Z aiperf010 stop begin
+2026-11-08T05:57:21Z aiperf010 stop took 120s: status=exited exit=137 restart=no
+2026-11-08T05:57:21Z qwen38-b-hr stop begin
+2026-11-08T05:57:27Z qwen38-b-hr stop took 6s: status=exited exit=0 restart=no
+2026-11-08T05:57:27Z qwen38-a-hr stop begin
+2026-11-08T05:57:31Z qwen38-a-hr stop took 4s: status=exited exit=0 restart=no
+2026-11-08T05:57:42Z vram after: card1 0.00;card2 0.00;card3 0.00;card4 0.00;
+2026-11-08T06:29:04Z qwen38-a-hr start rc=0
+2026-11-08T06:29:05Z qwen38-b-hr start rc=0
+2026-11-08T06:29:05Z aiperf010 start rc=0
+2026-11-08T06:30:05Z OK qwen38-a-hr port=30000 http=200 total_s=0.386094 {"model":"Qwen3.8-27B","content":"Paris","finish":"stop","usage":{"prompt_tokens":24,"total_tokens":26,"completion_tokens":2,"prompt_tokens_details":null,"reasoning_tokens":0}} after 60s
+2026-11-08T06:30:06Z OK qwen38-b-hr port=30001 http=200 total_s=0.374647 {"model":"Qwen3.8-27B","content":"Paris","finish":"stop","usage":{"prompt_tokens":24,"total_tokens":26,"completion_tokens":2,"prompt_tokens_details":null,"reasoning_tokens":0}} after 0s
+2026-11-08T06:30:11Z aiperf010 status=running started=2026-11-08T06:29:05.053702422Z restart=no
+2026-11-08T06:30:16Z vram after restore: card1 26.46;card2 15.09;card3 26.46;card4 15.09;
+2026-11-08T06:30:16Z restore rc=0
+aiperf010
+qwen38-b-hr
+qwen38-a-hr
diff --git a/canary/README.md b/canary/README.md
index bf00314f..fc5af0e7 100644
--- a/canary/README.md
+++ b/canary/README.md
@@ -13,10 +13,27 @@ secrets.
 
 | Kind | Engines | What it proves |
 |---|---|---|
-| runtime | `vllm-cpu`, `ovms-cpu`, `llama-cpu`; `eat-vllm-cpu` (informational) | The pinned image serves a pinned Qwen3-0.6B the way the launcher runs it. It returns one completion (HTTP 200, `finish_reason` stop or length, tokens > 0, the text matches `/Paris/`) and one stream (at least 2 chunks, then `[DONE]`), and the server reports the version its tag names. |
+| runtime | `vllm-cpu`, `ovms-cpu`, `llama-cpu`; `eat-vllm-cpu` (informational) | The pinned image serves a pinned Qwen3-0.6B under `docker run`, with the arguments and environment `engines.toml` gives it (the launcher's, or the fit's for vLLM on the CPU), not in the launcher's pod. It returns one completion (HTTP 200, `finish_reason` stop or length, tokens > 0, the text matches `/Paris/`) and one stream (at least 2 chunks, then `[DONE]`), and the server reports the version its tag names. |
 | surface | `vllm-xpu`, `llm-scaler`, `gaudi-vllm`, `llama-sycl`, `llama-vulkan`, `llama-full`, `dlstreamer`, `sglang-xeon`, `npu-umd`, `ovms` (and the runtime engines first) | Every flag and env name the launcher and `console/model_opt.py` emit (`canary/flags.sh`) exists in the image, and the constants derived from it still match: `console/plan.py`'s vLLM versions, and the NPU driver and firmware date. No device is needed. |
 | node | `node` | The root Dockerfile builds. `k3d cluster create --image` it with the k3d `packaging/install.sh` pins and the arguments the launcher passes on a host without NVIDIA (no load balancer, no traefik, the node labelled `nvidia.com/gpu.present=false`). Within 5 minutes the node is Ready, every baked manifest is in place, every DaemonSet they define exists, and every pod in every namespace (kube-system, NFD, the device plugins) is Running and ready or Succeeded; then the pods hold for 20 s with no container restarting. A plugin that needs hardware gets no pod on a hosted runner (its node selector), so a pod still Pending at the timeout, such as one whose image does not pull, fails. |
 
+## What it does not cover
+
+The runtime check runs the image, not the launcher's pod. Kubernetes adds nothing to its
+environment (no Service link variables such as `<SERVICE>_PORT=tcp://...`), and
+`probe.py` sends its own requests with `Content-Type: application/json`, not the
+launcher's warm-up. A fault in the pod around the engine, or in how the launcher asks
+it, therefore passes. The two faults that kept the launcher's vLLM agent from serving in
+every release up to 0.1.93 were of that kind: its Service, named `vllm`, gave the pod
+`VLLM_PORT=tcp://...`, on which vLLM exits, and its warm-up reached vLLM with no
+`Content-Type`, which vLLM answers with 400
+([#339](https://github.com/88plug/k3d-gpu/pull/339), fixed in 0.1.94). The images that
+agent serves (`vllm-xpu`, `llm-scaler`, `gaudi-vllm`) are surface engines only: a hosted
+runner has no Intel GPU or Gaudi, so the canary never serves them. `tests/detect_test.sh`
+checks the rendered pod spec and the warm-up command against stubs; nothing in CI serves
+through them. The fix was checked by hand on an Arc Pro B70, for `vllm-xpu` only
+([record](../bench/vllm-agent/results/2026-11-08/README.md)).
+
 ## Files
 
 | File | Role |
diff --git a/docs/first-run.md b/docs/first-run.md
index 3fb78fca..17cf5990 100644
--- a/docs/first-run.md
+++ b/docs/first-run.md
@@ -160,9 +160,11 @@ Predicted: 20 to 40 min.
 
 What differs: **the LLM runs on the CPU, not the NVIDIA GPU.** The cluster
 advertises `nvidia.com/gpu` for your own pods, but `agent up` and quickstart
-serve on Gaudi, Intel XMX GPUs or the CPU only. To serve an LLM on the NVIDIA
-GPU, see [GPUStack](integrations/gpustack.md) Recipe A; it is not verified on a
-k3d-gpu cluster.
+serve on Intel XMX GPUs or the CPU only (their Gaudi path, vLLM, does not serve in
+any release up to 0.1.93 and has not yet run on Gaudi hardware:
+[status](../README.md#vllm-on-gaudi)). To serve an LLM on
+the NVIDIA GPU, see [GPUStack](integrations/gpustack.md) Recipe A; it is not verified
+on a k3d-gpu cluster.
 
 Predicted: 15 to 30 min. The single measured run (H9, 303 s) was on a host of
 this class, with the LLM on 16 P-cores.
diff --git a/docs/fit.md b/docs/fit.md
index e9def9cf..ff8507f6 100644
--- a/docs/fit.md
+++ b/docs/fit.md
@@ -199,8 +199,11 @@ and `K3D_GPU_AGENT_GPUS` are inputs the fit checks, not values that replace it.
 or, with it unset, an OpenVINO IR): an OVMS pod holds one GPU, and `fit --gpus 2` with OVMS
 is refused as "OVMS serves one GPU". With
 `K3D_GPU_AGENT_BACKEND` unset, the engine the fit picks serves the model: OVMS, vLLM or
-llama-server. With `--model`, the model optimizer first picks the variant for the default
-engine (OVMS: Gemma 4's OpenVINO IR, for example). When that engine has nothing to serve,
+llama-server (vLLM does not serve in any release up to 0.1.93; release 0.1.94 served
+it end to end on an Arc Pro B70, and on a Gaudi it has not yet run on hardware:
+[status](../README.md#vllm-backends-for-discrete-arc-battlemage)). With `--model`, the
+model optimizer first picks the variant for the default engine (OVMS: Gemma 4's OpenVINO
+IR, for example). When that engine has nothing to serve,
 as with Qwen3.8-Flash-Next, the id is served as given, by the engine its fit picks. When
 the fit then picks another engine, the optimizer's choices are made again for that engine,
 pinned to the model the fit sized. A backend that is set and cannot serve the fit is
diff --git a/docs/intel.md b/docs/intel.md
index 33f68d29..0820b511 100644
--- a/docs/intel.md
+++ b/docs/intel.md
@@ -235,7 +235,7 @@ and installs through EAS.
 | OS | Ubuntu 22.04 / 24.04, x86_64 | Any Linux with Docker that runs k3d; `host-setup` prints pinned driver steps for Ubuntu 24.04/26.04, Fedora, Arch and Manjaro |
 | Cluster setup | Kubespray onto the host, or brownfield on an existing cluster; the `ei-*-eligible` labels come only from the Kubespray path | One `k3d-gpu up`; `K3D_GPU_NODE_LABELS` sets EAT's labels, `K3D_GPU_BALLOONS=external` hands the NRI policy to EAT [ledger:H14] |
 | EAT 2.0 install | EAS `es_auto_installer.sh install agent-toolkit`, onto Kubespray or an existing cluster (`existing_kubernetes`) | `k3d-gpu eat up` (`eas up --agent-toolkit`): the existing-cluster mode at the GA tags (EAS `v1.2.0`, EAT `release-2.0.0`), with the k3d settings and patches chosen from the cluster ([intel/eas.md](intel/eas.md)). F5 checks it on the Xeon 6741P. F3 (EAS `main`) is REFUTED ([run](../EXPERIMENTS.md#f3-run-i9-13900t-2026-11-01)): it installed, checked and tore down, and the secret scan found the Keycloak admin password on a command line |
-| Engines | vLLM CPU (`vllm/vllm-openai-cpu:v0.24.0`, bf16), 9 model profiles in the single-node table (7 chat, 1 embedding, 1 reranker). An OVMS Helm chart ships but no default path deploys it: `openvino/model_server:2025.4`, plus an unpinned `latest-py` tag for its model-export containers | `agent up` picks per host: vLLM on a Gaudi, OVMS 2026.4.1 on an XMX GPU, else OVMS on the P-cores, or, on a CPU class where an engine-choice row is SUPPORTED (the Xeon 6741P today), llama.cpp with the model's int4 GGUF (for Gemma 4 26B-A4B, the one model measured, Google's QAT Q4_0 GGUF and QAT MTP drafter) [ledger:H59]; vLLM XPU (`v0.32.0`) or llm-scaler on Arc by flag. Its images are pinned by digest in `images.lock` |
+| Engines | vLLM CPU (`vllm/vllm-openai-cpu:v0.24.0`, bf16), 9 model profiles in the single-node table (7 chat, 1 embedding, 1 reranker). An OVMS Helm chart ships but no default path deploys it: `openvino/model_server:2025.4`, plus an unpinned `latest-py` tag for its model-export containers | `agent up` picks per host: vLLM on a Gaudi, OVMS 2026.4.1 on an XMX GPU, else OVMS on the P-cores, or, on a CPU class where an engine-choice row is SUPPORTED (the Xeon 6741P today), llama.cpp with the model's int4 GGUF (for Gemma 4 26B-A4B, the one model measured, Google's QAT Q4_0 GGUF and QAT MTP drafter) [ledger:H59]; vLLM XPU (`v0.32.0`) or llm-scaler on Arc by flag. No release up to 0.1.93 serves with vLLM. Release 0.1.94 served vLLM XPU end to end on an Arc Pro B70; vLLM on a Gaudi and llm-scaler have not yet run on hardware ([status](../README.md#vllm-backends-for-discrete-arc-battlemage)). Its images are pinned by digest in `images.lock` |
 | Accelerators | None requested: no `gpu.intel.com`, `habana.ai` or `/dev/dri` on the default path | Intel GPU, NPU, Gaudi, DSA, IAA and QAT plugins started per detected device, and `up` checks that each is schedulable (on the Xeon 6741P it warned when QAT advertised none [ledger:H52b]) |
 | CPU placement | NRI balloons sized for Xeon NUMA; on a hybrid CPU EAT's reservation needs the patch in [`contrib/eat/`](../contrib/eat) | Balloons baked for hybrid P/E and multi-NUMA hosts; vLLM on P-cores and agent services on E-cores through `contrib/eat/` values |
 
@@ -511,12 +511,17 @@ has not run on that host yet.
 
 Shipped since this roadmap was first written:
 
-- **Agent endpoint.** `k3d-gpu agent up`: an OpenAI-compatible server that picks vLLM on
-  a Gaudi, OVMS on an XMX Intel GPU, else OVMS on the P-cores, from what the node
-  advertises; vLLM XPU or llm-scaler on Arc by flag ([README](../README.md)). On a CPU
-  class where an engine-choice row is SUPPORTED, llama.cpp with the model's int4 GGUF takes
-  the CPU instead: the Xeon 6741P class since 2026-11-01, measured on Gemma 4 26B-A4B with
-  the QAT GGUF and MTP drafter only [ledger:H59].
+- **Agent endpoint.** `k3d-gpu agent up`: an OpenAI-compatible server that picks OVMS on
+  an XMX Intel GPU, else OVMS on the P-cores, from what the node advertises; vLLM XPU on
+  Arc by flag ([README](../README.md)). On a CPU class where an engine-choice row is
+  SUPPORTED, llama.cpp with the model's int4 GGUF takes the CPU instead: the Xeon 6741P
+  class since 2026-11-01, measured on Gemma 4 26B-A4B with the QAT GGUF and MTP drafter
+  only [ledger:H59]. vLLM XPU serves from release 0.1.94: it served end to end on an Arc
+  Pro B70 on 2026-11-08 with `vllm/vllm-openai-xpu:v0.32.0` and `Qwen/Qwen3-4B`
+  ([record](../bench/vllm-agent/results/2026-11-08/README.md)). No release up to 0.1.93
+  serves with vLLM. Its other vLLM paths (vLLM on a Gaudi by default, llm-scaler on Arc by
+  flag) are in the launcher but have not yet run on hardware; see the vLLM agent item
+  under Planned below.
 - **EAS and EAT 2.0.** `k3d-gpu eas up`, `status`, `check` and `down` (and `k3d-gpu eat`,
   the agent-toolkit layer) drive Intel's EAS installer in its bring-your-own-cluster mode,
   at the GA tags of 2026-11-06, with every k3d setting and patch chosen from the cluster
@@ -562,6 +567,16 @@ Shipped since this roadmap was first written:
 Planned, not shipped. Each item lands with its own test and, where it touches
 hardware, a recorded run.
 
+- **vLLM agent on a Gaudi and llm-scaler.** In releases 0.1.21 through 0.1.93, `agent up`
+  with vLLM does not serve, on Arc or on a Gaudi. The hardware canary for the `v0.33.0`
+  bump found why on an Arc Pro B70 on 2026-11-07. The agent's Service is named `vllm`, so
+  Kubernetes gives the pod `VLLM_PORT=tcp://...`; vLLM reads that as its own port and
+  exits. The warm-up request carries no `Content-Type`, and vLLM answers it with 400
+  ([README](../README.md#vllm-backends-for-discrete-arc-battlemage)).
+  [#339](https://github.com/88plug/k3d-gpu/pull/339) fixed both in release 0.1.94, whose
+  vLLM XPU path served end to end on an Arc Pro B70 on 2026-11-08
+  ([record](../bench/vllm-agent/results/2026-11-08/README.md)). The Gaudi path and
+  llm-scaler get the same fix and have not yet run on hardware.
 - **Hardware coverage.** Move the `xe`, NPU, Gaudi, DSA, IAA, QAT and AMX rows to
   verified on real Core Ultra, Arc, Gaudi 3 and Xeon hosts. The campaign on a Xeon 6741P
   with 4x Arc Pro B70 closed on 2026-11-01 ([above](#eat-by-default-vs-with-k3d-gpu)). Its
diff --git a/docs/intel/upstream-eat.md b/docs/intel/upstream-eat.md
index 3eb5a985..36108d34 100644
--- a/docs/intel/upstream-eat.md
+++ b/docs/intel/upstream-eat.md
@@ -75,7 +75,7 @@ cluster, read from source only:
 
 | #21 adds | Already covered by k3d-gpu | Expected on a k3d-gpu cluster |
 |---|---|---|
-| Arc B60/B70 XPU path (`compute_platform=xpu`, `vllm-openai-xpu:v0.28.0`, TP4 `gemma-4-31B-it`) | B70 detection, `gpu.intel.com/xe` asserted by `up`, vLLM XPU `v0.32.0` by digest through `agent up` | Host Level Zero and OpenCL libraries mounted by `hostPath` resolve inside the k3d node container, not on the host |
+| Arc B60/B70 XPU path (`compute_platform=xpu`, `vllm-openai-xpu:v0.28.0`, TP4 `gemma-4-31B-it`) | B70 detection, `gpu.intel.com/xe` asserted by `up`, vLLM XPU `v0.32.0` by digest through `agent up`, served end to end on an Arc Pro B70 at release 0.1.94; no release up to 0.1.93 serves it ([status](../../README.md#vllm-backends-for-discrete-arc-battlemage)) | Host Level Zero and OpenCL libraries mounted by `hostPath` resolve inside the k3d node container, not on the host |
 | Intel GPU plugin from intel-device-plugins `main` (`intel_gpu_plugin_ref`, unpinned) | Plugin `0.37.1` baked into the node image | Two GPU plugin DaemonSets on one node |
 | NUMA-aware NRI balloons, one `BalloonsPolicy` per node | Baked balloons for hybrid P/E and multi-NUMA hosts; `K3D_GPU_BALLOONS=external` [ledger:H14] | Two policies if `deploy_nri_balloon_policy=yes`; `eat-up.sh` sets `off` |
 | Kata sandboxes (`kata-fc`, `kata-qemu`) | `doctor --perf` reports `/dev/kvm` on the host | Kata needs `/dev/kvm` inside the k3d node, which `up` does not pass through |
diff --git a/docs/platforms.md b/docs/platforms.md
index 9d94c161..63a18236 100644
--- a/docs/platforms.md
+++ b/docs/platforms.md
@@ -394,12 +394,15 @@ B70 runs land:
   `/usr/lib/x86_64-linux-gnu/intel-opencl`).
 - What k3d-gpu already covers: B70 detection and the Intel GPU plugin `0.37.1`, baked
   and pinned; `gpu.intel.com/xe` advertised and asserted by `up`; the label through
-  `K3D_GPU_NODE_LABELS`; vLLM XPU through `K3D_GPU_AGENT_BACKEND=vllm` (`v0.32.0`, by digest).
+  `K3D_GPU_NODE_LABELS`; vLLM XPU through `K3D_GPU_AGENT_BACKEND=vllm` (`v0.32.0`, by
+  digest) from release 0.1.94, which served it end to end on an Arc Pro B70
+  ([record](../bench/vllm-agent/results/2026-11-08/README.md)). No release up to 0.1.93
+  serves with it ([status](../README.md#vllm-backends-for-discrete-arc-battlemage)).
 - Expected collision: a second GPU plugin. `deploy-intel-gpu-plugin.yml` applies
   `deployments/gpu_plugin/base/intel-gpu-plugin.yaml` from intel-device-plugins `main`
   (`intel_gpu_plugin_ref` defaults to `main`, unpinned) beside the baked DaemonSet.
   `eat-up.sh` therefore stops on `compute_platform=xpu` and points at
-  `K3D_GPU_AGENT_BACKEND=vllm`.
+  `K3D_GPU_AGENT_BACKEND=vllm` with k3d-gpu 0.1.94 or later.
 - Expected collision: a second balloons policy. The reworked `nri_cpu_balloons` role
   renders one NUMA-sized `BalloonsPolicy` per node (`node.<name>`). #21 still installs
   it only on `deploy_nri_balloon_policy == "yes"` (`fresh-install.sh` line 247), so
diff --git a/docs/quickstart.md b/docs/quickstart.md
index 16869603..ac414fad 100644
--- a/docs/quickstart.md
+++ b/docs/quickstart.md
@@ -67,6 +67,9 @@ You see five steps, each with a progress bar:
    used for this LLM, and neither is an iGPU without XMX (UHD, Iris Xe,
    N-series). On a Gaudi the server is vLLM instead, and the
    chat UI step needs `--no-ui` (the UI talks to OpenVINO Model Server only).
+   In releases up to 0.1.93 this step does not finish on a Gaudi: they do not
+   serve with vLLM. Release 0.1.94 served vLLM end to end on an Arc Pro B70; it
+   has not yet run on Gaudi hardware ([status](../README.md#vllm-on-gaudi)).
 4. **Chat UI**: Open WebUI, with an admin account created for you.
 5. **Port-forward**: the UI on `127.0.0.1` only, not on your network.
 
diff --git a/integrations/eat/README.md b/integrations/eat/README.md
index c3d42fac..1c82cb5f 100644
--- a/integrations/eat/README.md
+++ b/integrations/eat/README.md
@@ -301,8 +301,11 @@ except where an item says otherwise, and item 11 is new there. In
     stops before touching the cluster, reading each key as `read_config_file`
     does, on `compute_platform=xpu` (fresh-install.sh:219 applies
     intel-device-plugins' `gpu_plugin` from `main`, unpinned, beside the plugin
-    k3d-gpu bakes; vLLM on an Intel GPU is `K3D_GPU_AGENT_BACKEND=vllm k3d-gpu
-    agent up`), `deploy_kata_containers=on` (Kata needs `/dev/kvm`, which a k3d
+    k3d-gpu bakes; the message points at k3d-gpu's vLLM on an Intel GPU,
+    `K3D_GPU_AGENT_BACKEND=vllm k3d-gpu agent up`, with k3d-gpu 0.1.94 or later: no
+    release up to 0.1.93 serves with it, and 0.1.94 served it end to end on an Arc Pro
+    B70, see [the status](../../README.md#vllm-backends-for-discrete-arc-battlemage)),
+    `deploy_kata_containers=on` (Kata needs `/dev/kvm`, which a k3d
     node lacks) and `rack_scale_mode=on` (it builds its own cluster from
     `rack-profile.yaml`: netplan, BGP, the NVIDIA GPU operator). None of these has
     run here; each is read from source.
diff --git a/integrations/eat/eat-up.sh b/integrations/eat/eat-up.sh
index 4e5ed978..bbf4f6ce 100755
--- a/integrations/eat/eat-up.sh
+++ b/integrations/eat/eat-up.sh
@@ -82,7 +82,7 @@ case "$(cfg_value "${EAT_CONFIG}" deploy_nri_balloon_policy)" in
     *) err "${EAT_CONFIG}: deploy_nri_balloon_policy must be off: ${EAT_TAG} deploys its own NRI balloons policy when it is on or absent, beside k3d-gpu's, which must remain the only one, and in model mode it moves vLLM to a balloon type k3d-gpu's policy does not define" ;;
 esac
 case "$(cfg_value "${EAT_CONFIG}" compute_platform)" in
-    x|xpu) err "compute_platform=xpu: ${EAT_TAG} then applies intel-device-plugins' gpu_plugin from its main branch, unpinned, beside the GPU plugin k3d-gpu bakes; for vLLM on an Intel GPU run K3D_GPU_AGENT_BACKEND=vllm k3d-gpu agent up" ;;
+    x|xpu) err "compute_platform=xpu: ${EAT_TAG} then applies intel-device-plugins' gpu_plugin from its main branch, unpinned, beside the GPU plugin k3d-gpu bakes; for vLLM on an Intel GPU run K3D_GPU_AGENT_BACKEND=vllm k3d-gpu agent up with k3d-gpu 0.1.94 or later (earlier releases do not serve it)" ;;
 esac
 case "$(cfg_value "${EAT_CONFIG}" deploy_kata_containers)" in
     on|yes) err "deploy_kata_containers=on: Kata needs /dev/kvm on the node, and a k3d node is a container without it; set deploy_kata_containers=off" ;;
diff --git a/pins/classes b/pins/classes
index a288319f..e0f8f07c 100644
--- a/pins/classes
+++ b/pins/classes
@@ -34,6 +34,7 @@ oracle/seed/**                          historical  the maintainer bundles of th
 oracle/raw/**                           historical  the raw H6 and H8 records the seed bundles were built from
 bench/f3/**                             historical  F3's runner and records (a functional row; ledger lines take H ids only)
 bench/f5/**                             historical  F5's runner and records: its pins are the pre-registration's (a functional row; ledger lines take H ids only)
+bench/vllm-agent/**                     historical  the vLLM agent's unpatched end-state runs on an Arc Pro B70 (2026-11-08): release 0.1.94 and #357's head
 
 # --- Runners of finished rows. Ledger lines, so the class follows the ledger: every row
 # listed has a verdict, so each resolves to historical (hack/pin-classes effective).
diff --git a/tests/eat_test.sh b/tests/eat_test.sh
index 595f4855..113702e6 100755
--- a/tests/eat_test.sh
+++ b/tests/eat_test.sh
@@ -700,8 +700,16 @@ cfg="${WORK}/xpu-reason.cfg"
 cp "${EAT}/agentic-config.cfg" "${cfg}"
 echo 'compute_platform=xpu' >> "${cfg}"
 out=$(EAT_CONFIG="${cfg}" EAT_KUBECONFIG=/nonexistent "${EAT}/eat-up.sh" 2>&1) || true
-if grep -qF 'K3D_GPU_AGENT_BACKEND=vllm' <<< "${out}"; then ok "eat-up: the xpu refusal points at vLLM on an Intel GPU through k3d-gpu agent up"
+if grep -qF 'K3D_GPU_AGENT_BACKEND=vllm' <<< "${out}"; then ok "eat-up: the xpu refusal names k3d-gpu's vLLM on an Intel GPU"
 else fail "eat-up: the xpu refusal names no alternative: ${out}"; fi
+# Releases through 0.1.93 never served that path. Release 0.1.94 (#339) is the first that
+# does, and it served end to end on an Arc Pro B70: the refusal sends users there and
+# names that release, since an older launcher does not serve.
+if grep -qF 'K3D_GPU_AGENT_BACKEND=vllm k3d-gpu agent up' <<< "${out}" \
+    && grep -qF 'k3d-gpu 0.1.94 or later' <<< "${out}" \
+    && ! grep -qF 'has not yet served' <<< "${out}"; then
+    ok "eat-up: the xpu refusal sends users to k3d-gpu's vLLM on an Intel GPU, from release 0.1.94"
+else fail "eat-up: the xpu refusal does not send users to k3d-gpu's vLLM from release 0.1.94: ${out}"; fi
 
 # 5. The GenAI Gateway path installs pinned charts and images. EAT's
 #    deploy-genai-gateway.yml syncs helm_charts_base_genai{,_trace} to the node, runs