Skip to content

Update to ort 1.3.0.0 and genai to 0.16.0 - #1102

Merged
Baiju Meswani (baijumeswani) merged 5 commits into
mainfrom
baijumeswani/update-ort-1.30-genai-0.16
Sep 14, 2026
Merged

Baiju Meswani (baijumeswani) merged 5 commits into
mainfrom
baijumeswani/update-ort-1.30-genai-0.16

Conversation

@baijumeswani

Copy link
Copy Markdown
Collaborator

No description provided.

Copilot AI balanced review requested due to automatic review settings September 13, 2026 08:30
@vercel

vercel Bot commented Sep 13, 2026

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated
foundry-local Ready Ready Preview Sep 14, 2026 4:38pm UTC

Request Review

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🟡 Changes recommended

Unconditional Engine compilation can break older or local GenAI configurations; retain capability gating or add minimum-version validation.

Get a fresh assessment by requesting another Copilot review.

Pull request overview

Updates the SDK to ONNX Runtime 1.30.0 and GenAI 0.16.0, refreshing Engine, CUDA, and CI configuration.

Changes:

  • Synchronizes dependency pins across SDKs, packaging, and pipelines.
  • Enables Engine sources and tests unconditionally.
  • Refreshes CUDA EP manifests, hashes, and Engine CI settings.
File summaries
File Summary
sdk_v2/rust/deps_versions.json Synchronizes Rust dependency pins.
sdk_v2/js/test/install-native.test.ts Updates installer version expectations.
sdk_v2/deps_versions.json Updates canonical dependency pins.
sdk_v2/cpp/test/internal_api/cuda_ep_bootstrapper_test.cc Refreshes CUDA manifest expectations.
sdk_v2/cpp/test/CMakeLists.txt Includes Engine tests unconditionally.
sdk_v2/cpp/src/inferencing/generative/genai_model_instance.h Removes conditional Engine declarations.
sdk_v2/cpp/src/inferencing/generative/genai_model_instance.cc Enables unconditional Engine initialization.
sdk_v2/cpp/src/inferencing/generative/chat/chat_session.cc Removes Engine availability guards.
sdk_v2/cpp/src/ep_detection/cuda_ep_manifest.cc Updates CUDA bundle metadata and hashes.
sdk_v2/cpp/CMakeLists.txt Builds Engine sources unconditionally; older or local GenAI overrides may fail to compile.
.pipelines/v2/templates/stages-test-engine.yml Updates Engine CI dependencies.
.pipelines/v2/templates/stages-sdk-v2.yml Passes the GenAI version to Engine tests.
.pipelines/v2/sdk_v2-pipeline-plan.md Documents updated dependency versions.
.pipelines/foundry-local-packaging.yml Updates packaging dependency variables.
.github/instructions/cpp-ci-test-policy.instructions.md Updates Engine CI policy.
Review details
  • Files reviewed: 15/15 changed files
  • Comments generated: 1
  • Review effort level: Lite (auto)

Note

Copilot is running an experiment and ran this review at Lite.


💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

Comment thread sdk_v2/cpp/CMakeLists.txt
bmehta001
bmehta001 previously approved these changes Sep 14, 2026
@baijumeswani
Baiju Meswani (baijumeswani) merged commit 97068a8 into main Sep 14, 2026
61 checks passed
@baijumeswani
Baiju Meswani (baijumeswani) deleted the baijumeswani/update-ort-1.30-genai-0.16 branch September 14, 2026 18:12
Baiju Meswani (baijumeswani) added a commit that referenced this pull request Sep 14, 2026
## Summary

- treat `temperature: 1` as neutral when `do_sample` is omitted
- keep the model package default authoritative
- preserve existing behavior for non-neutral temperatures and explicit
`do_sample` values

## Why

OpenAI-compatible coding clients such as GitHub Copilot CLI send
`temperature: 1` even when they are not requesting a sampling-policy
override. Foundry previously translated every positive temperature into
`do_sample=true`. That replaced the model package default and disabled
DFlash2 drafting for packages whose default is `do_sample=false`.

ORT GenAI already owns speculative-decoding eligibility: greedy turns
may draft, while sampled turns run target-only. Foundry therefore does
not need dflash2-specific detection, rejection logic, public controls,
or test-only Engine statistics APIs.

## Behavior

- omitted `do_sample`, omitted temperature: preserve all model defaults
- omitted `do_sample`, `temperature: 1`: forward the neutral temperature
and preserve the model default
- omitted `do_sample`, non-neutral positive temperature: continue
resolving to `do_sample=true`
- explicit `do_sample=true`: remains explicit, including with
`temperature: 1`

## Validation

- focused `SamplingPlanTest.*` suite passes
- real dflash2 release qualification confirmed that default, explicit
greedy, and neutral-temperature turns draft
- sampled turns complete through ORT GenAI target-only with zero draft
activity
- integrated Toolkit typed/reverse-result gate passes
- stock GHCP coding gate proceeds with the package-owned greedy default
and speculative decoding enabled

Stacked on #1102 for ORT 1.30.0 and ORT GenAI 0.16.0.

---------

Copilot-Session: 6883972e-cc0b-4076-a3a5-86106e7a1133

This branch was successfully deployed

1 active deployment
Preview 57d17609 Deployed Sep 14, 2026 by vercel[bot]
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants