FonaTech · Nov 11, 2026 · Nov 11, 2026
diff --git a/CHANGELOG.md b/CHANGELOG.md
diff --git a/README.md b/README.md
@@ -113,13 +113,7 @@ At this size a full-depth step costs about as much as a batched verify pass. Wit
 
 `chronos.model.factory.fit_config` searches for a configuration that matches total and active parameter targets.
 
-**Kernel and extensions.** The `chronos` package trains general models and carries no task-specific code. Task tooling lives in extensions that import the kernel. The en↔zh translation demo is one: `demos/translation` (`chronos-translate`: corpus pipeline, BLEU/chrF runs, model bundles; see its README). It is not part of the `Project_Chronos` wheel and is not published on PyPI; install it from a checkout with `pip install -e demos/translation`.
-
-**Status of the 50M en→zh demo (in progress; `demos/translation`).**
-- Model: 49.69M total / 12.09M active parameters, 36 experts, MLA latent 128, SWA window 128, 16k self-built vocabulary.
-- Run: a 72-hour MLX bf16 run (pretrain, then SFT) started on 2026-11-05.
-- Interim measurement: valid CE 2.477 at pretrain step 13,996 of about 24.7k planned (it was 3.012 at step 1,656). That checkpoint is packaged as `chronos-enzh-50m-pretrain-s13996` (see Model bundle).
-- BLEU/chrF on FLORES+ and WMT22, and the ablations (attention mode, depth, lazy-expert budget, speculative γ), will be published after the run finishes. No quality numbers are claimed before then.
+**Kernel and extensions.** The `chronos` package trains general models and carries no task-specific code. Task tooling lives in extensions that import the kernel. The en↔zh translation demo is one: `demos/translation` (`chronos-translate`: corpus pipeline, BLEU/chrF runs, model bundles; see its README). It is not part of the `Project_Chronos` wheel and is not published on PyPI; install it from a checkout with `pip install -e demos/translation`. The 50M en→zh status, the released model `chronos-enzh-50m-sft` with its measurements, and the model bundle are documented in [`demos/translation/README.md`](demos/translation/README.md).
 
 **Inference and memory options (all off by default).** Each option is a config field and a `chronos serve` / Inference-tab override. Tests compare each one with its reference path on Torch and MLX.
 
@@ -501,22 +495,6 @@ All lambda terms are searchable with Optuna TPE, together with structural hyperp
 
 ---
 
-## Released model: chronos-enzh-50m-sft
-
-[`models/chronos-enzh-50m-sft/`](models/chronos-enzh-50m-sft/) packages a 49.7M-parameter Chronos v3 model (12.1M active per token) trained for 69 hours on one Apple M5 Max: configuration, tokenizer, trained prompt, decoding settings, a driver (`translate.py`) and the evaluation behind every number. The weights (99.5 MB, bfloat16 safetensors) are attached to the release [`enzh-50m-sft-v1`](https://github.com/FonaTech/Project_Chronos/releases/tag/enzh-50m-sft-v1).
-
-- **Boundary:** English → Chinese only, one sentence (≤ 128 tokens) at a time. Chinese → English is disabled: English-only fine-tuning erased it (FLORES+ BLEU 0.00 after SFT). The driver refuses input outside the boundary.
-- **Measured** (MLX bf16, greedy): FLORES+ devtest BLEU 33.48 / chrF 29.52; WMT22 BLEU 32.60 / chrF 28.83; WMT24++ BLEU 26.89 / chrF 24.43 (split into sentences).
-- **Research use only.** The training corpora's licences are unknown; see the folder's `NOTICE`.
-
-```bash
-cd models/chronos-enzh-50m-sft
-pip install --pre "Project_Chronos[mlx]==0.3.2a0"   # or "Project_Chronos==0.3.2a0" without MLX
-python translate.py --download && python translate.py "The meeting was moved to Friday."
-```
-
----
-
 ## Installation
 
 ```bash
@@ -598,6 +576,7 @@ The old preset names (Recommended-CN, Tiny/Small/Medium/Large) still load; they
 3. **Sets LR and batch size** from the DeepSeek LLM compute-optimal fits (η = 0.3118·C^−0.125, B = 0.2920·C^0.3271), capped for wide models.
 4. **Fits memory.** It picks micro-batch, accumulation, int8 KV, router scope, activation checkpointing, optimizer paging and an expert RAM budget for the memory level.
 5. **Writes per-stage values.** Pretrain, SFT, DPO, ORPO, GRPO and distill each get their own learning rate, batch, epochs, save interval and stage-specific terms (β, temperature, α).
+6. **Sets the loop knobs (depth above 1).** `recurrent_order="block"`, `recurrent_dt_param="sigmoid"` and `recurrent_dt_init = min(0.5, 2 / depth)`, each with a `sources` entry in the report (the first two cite the 15M measurements; the step size is labelled estimated). One loop iteration is an Euler step of `dh/dτ = block(h) − h`, so `depth × dt` is the integration horizon; the planner holds it at 2 and lets depth refine the integration. Kernel defaults are unchanged (`layer`, `clamp`, 0.25). All of this was measured on one seed per arm at 15M parameters; see the CHANGELOG, "Physical mechanism" and "Known limits".
 
 The stage table is editable; the Train tab merges the selected stage's row over the shared values, and the pretrain row stays in sync with the shared widgets. "Save plan" writes the config, stage values and report to JSON. Every number in the report is labelled measured or estimated.
 
@@ -646,38 +625,6 @@ OLLAMA_HOST=127.0.0.1:11435 ollama run <model-name> "Hello"
 - Stock `ollama` / llama.cpp cannot execute Chronos GGUF files (MLA+SWA hybrid layers and the contractive recurrence have no stock equivalent); use this server instead.
 - Security: binds 127.0.0.1 without authentication by default. Other addresses require `--allow-remote` and `--api-key`; clients then send `Authorization: Bearer <key>`.
 
-### Model bundle (translation demo extension)
-
-Bundles belong to the translation extension (`demos/translation`). Build the two wheels first (`python -m build --outdir dist .` and `python -m build --outdir dist demos/translation`). `chronos-translate bundle build` packages one exported checkpoint of a run into a folder that works right after unpacking. Building only reads the run directory.
-
-```bash
-chronos-translate bundle build --run-dir runs/enzh_50m_v1 --which best --pack \
-    --wheel dist/project_chronos-0.3.2a0-py3-none-any.whl --wheel dist/chronos_translate-0.3.2a0-py3-none-any.whl
-```
-
-| Bundle path | Content |
-|---|---|
-| `model/` | weights + `.config.json` sidecar, tokenizer (`model/tokenizer/`) |
-| `eval/` | FLORES+ dev/devtest (CC BY-SA 4.0) for `bundle eval` / `probe` |
-| `data/sft_valid/` | the run's SFT validation shards |
-| `wheels/`, `bin/setup.sh`, `bin/chronos` | offline install of the kernel and extension wheels into `.venv`; launcher that sets `CHRONOS_BUNDLE` and sends `bundle …` to `chronos-translate` |
-| `recipes/source_run_spec.json`, `recommended.json` | the recipe that produced the weights; fine-tune tiers `smoke` / `quick` / `standard` / `full` |
-| `bundle.json` | provenance (step, tokens, valid CE), measured quality probe, SHA-256 of every file |
-
-On the recipient side:
-
-```bash
-bin/setup.sh                                              # venv + wheels + `bundle check`
-bin/chronos bundle translate "The weather is very nice today."
-bin/chronos bundle translate --backend mlx --speculative --file examples/translate_en.txt
-bin/chronos bundle eval --split devtest -n 200
-bin/chronos bundle prepare-data my_pairs.jsonl --out work/data      # FLORES+ is removed from your data
-bin/chronos bundle finetune --data work/data --run-dir work/ft --tier quick --mode sft
-bin/chronos bundle serve -- --port 11435
-```
-
-`--speculative` needs `--repetition-penalty 1.0` (its default), because the penalty changes the target distribution at every position. `finetune` writes a `chronos-translate run` spec with `init_checkpoint` set to the bundle weights and refuses a run directory that already holds a run. Bundles built from data with unknown licences (WMT19, OPUS-100) say in `NOTICE.md` that the weights are for research only.
-
 ### Deployment export
 
 ```bash
@@ -797,11 +744,15 @@ tuner.start(
         tune_num_experts=True,
         tune_num_shared_experts=True,
         tune_kv_latent_dim=True,
+        tune_recurrent_order=True,      # probes a looped model (probe_recurrent_depth, default 4)
+        tune_recurrent_dt_init=True,
     ),
     n_trials=20,
 )
 ```
 
+The probe trains on real tokens from `dataset_path` (up to 400k, encoded with the model's tokenizer, or the bundled 32k one) and falls back to random ids with a log line when the data cannot be read. The two loop knobs are a coarse screen: a few steps of a 4-layer model cannot rank step sizes that differ by less than the run-to-run noise. `recurrent_later_groups` is not searched, because it trades quality for cost and a loss-only search would always choose the full block.
+
 ---
 
 ## Project layout

diff --git a/README_zh.md b/README_zh.md
@@ -125,13 +125,7 @@ flowchart LR
 
 `chronos.model.factory.fit_config` 按总参数量和激活参数量目标搜索模型配置。
 
-**内核与扩展。** `chronos` 包用于训练通用模型，不含任何任务特化代码。任务工具放在调用内核的扩展里。英↔中翻译 demo 就是这样一个扩展：`demos/translation`（`chronos-translate`：语料流水线、BLEU/chrF 训练、模型包，详见其 README）。它不在 `Project_Chronos` wheel 里，也不发布到 PyPI，需从仓库检出后用 `pip install -e demos/translation` 安装。
-
-**50M 英译中 demo（进行中；`demos/translation`）**
-- 模型：总参数 49.69M，激活 12.09M；36 个专家；MLA latent 128；SWA 窗口 128；自研 16k 词表。
-- 训练：72 小时 MLX bf16 任务（先预训练，后 SFT），2026-11-05 启动。
-- 中途实测：预训练第 13,996 步（计划约 24.7k 步）valid CE 为 2.477（第 1,656 步时为 3.012）。该 checkpoint 已打包为 `chronos-enzh-50m-pretrain-s13996`（见“模型包”）。
-- FLORES+ 和 WMT22 上的 BLEU/chrF，以及各项消融（注意力模式、深度、懒加载专家预算、推测 γ），等训练结束后公布。在此之前不给出任何质量数字。
+**内核与扩展。** `chronos` 包用于训练通用模型，不含任何任务特化代码。任务工具放在调用内核的扩展里。英↔中翻译 demo 就是这样一个扩展：`demos/translation`（`chronos-translate`：语料流水线、BLEU/chrF 训练、模型包，详见其 README）。它不在 `Project_Chronos` wheel 里，也不发布到 PyPI，需从仓库检出后用 `pip install -e demos/translation` 安装。50M 英译中的进展、已发布模型 `chronos-enzh-50m-sft` 及其实测结果、模型包，均写在 [`demos/translation/README_zh.md`](demos/translation/README_zh.md)。
 
 **推理与内存选项（全部默认关闭）。** 每个选项既是 config 字段，也可作为 `chronos serve` 和 Inference Tab 的覆盖项；测试在 Torch 和 MLX 上把每个选项与参考路径做对比。
 
@@ -505,22 +499,6 @@ D_{\mathrm{KL}}
 
 ---
 
-## 已发布模型：chronos-enzh-50m-sft
-
-[`models/chronos-enzh-50m-sft/`](models/chronos-enzh-50m-sft/) 打包了一个 4,969 万参数的 Chronos v3 模型（每个 token 激活 1,209 万），在一台 Apple M5 Max 上训练了 69 小时。目录中有模型配置、分词器、训练时的 prompt、解码参数、驱动脚本（`translate.py`），以及每个数字背后的评测。权重（99.5 MB，bfloat16 safetensors）作为附件放在 release [`enzh-50m-sft-v1`](https://github.com/FonaTech/Project_Chronos/releases/tag/enzh-50m-sft-v1) 中。
-
-- **边界：** 只做英译中，每次一句（≤ 128 token）。中译英已屏蔽：只含英译中的微调抹掉了这一能力（SFT 后 FLORES+ BLEU 0.00）。驱动会拒绝边界以外的输入。
-- **实测**（MLX bf16，贪心解码）：FLORES+ devtest BLEU 33.48 / chrF 29.52；WMT22 BLEU 32.60 / chrF 28.83；WMT24++ BLEU 26.89 / chrF 24.43（分句）。
-- **仅供研究使用。** 训练语料的许可未知，见目录中的 `NOTICE`。
-
-```bash
-cd models/chronos-enzh-50m-sft
-pip install --pre "Project_Chronos[mlx]==0.3.2a0"   # 不用 MLX 时装 "Project_Chronos==0.3.2a0"
-python translate.py --download && python translate.py "The meeting was moved to Friday."
-```
-
----
-
 ## 安装
 
 ```bash
@@ -601,6 +579,7 @@ Config Tab 提供 5 档 v3 预设。它们都由规划器生成，存放在 `ui/
 3. **确定学习率和 batch。** 按 DeepSeek LLM 的计算最优拟合式（η = 0.3118·C^−0.125，B = 0.2920·C^0.3271）给出，宽模型的学习率另设上限。
 4. **匹配内存。** 按内存档位选择 micro-batch、梯度累积、int8 KV、路由范围、激活检查点、优化器分页和专家 RAM 预算。
 5. **生成各阶段参数。** pretrain、SFT、DPO、ORPO、GRPO、distill 各有独立的学习率、batch、epoch、保存间隔，以及各自的专用项（β、温度、α）。
+6. **设置循环参数（深度大于 1 时）。** `recurrent_order="block"`、`recurrent_dt_param="sigmoid"` 和 `recurrent_dt_init = min(0.5, 2 / depth)`，报告的 `sources` 中各有一条来源（前两项引用 15M 实测，步长标注为估算）。一次循环迭代是 `dh/dτ = block(h) − h` 的一个欧拉步，所以 `depth × dt` 是积分时程；规划器把时程固定为 2，让深度只负责把积分切得更细。内核默认值不变（`layer`、`clamp`、0.25）。以上结论都来自 15M 参数、每组一个种子的实测；见 CHANGELOG 的“Physical mechanism”和“Known limits”。
 
 阶段表可以直接编辑。Train Tab 会把所选阶段那一行合并到共享参数之上；pretrain 行与共享控件保持同步。“保存方案”把 config、各阶段参数和报告写成 JSON。报告里每个数字都标注实测或估算。
 
@@ -650,38 +629,6 @@ OLLAMA_HOST=127.0.0.1:11435 ollama run <模型名> "你好"
 - stock `ollama` / llama.cpp 无法执行 Chronos 的 GGUF 文件，因为 MLA+SWA 混合层和收缩式循环没有对应的 stock 实现；请改用本服务。
 - 安全：默认只监听 127.0.0.1，不带认证。绑定其他地址必须同时给出 `--allow-remote` 和 `--api-key`，客户端随后发送 `Authorization: Bearer <key>`。
 
-### 模型包（翻译 demo 扩展）
-
-模型包属于翻译扩展（`demos/translation`）。先构建两个 wheel（`python -m build --outdir dist .` 和 `python -m build --outdir dist demos/translation`）。`chronos-translate bundle build` 把某次运行导出的一个 checkpoint 打包成解压即用的目录。打包过程只读取运行目录。
-
-```bash
-chronos-translate bundle build --run-dir runs/enzh_50m_v1 --which best --pack \
-    --wheel dist/project_chronos-0.3.2a0-py3-none-any.whl --wheel dist/chronos_translate-0.3.2a0-py3-none-any.whl
-```
-
-| 包内路径 | 内容 |
-|---|---|
-| `model/` | 权重 + `.config.json` 配置文件，分词器（`model/tokenizer/`） |
-| `eval/` | FLORES+ dev/devtest（CC BY-SA 4.0），供 `bundle eval` / `probe` 使用 |
-| `data/sft_valid/` | 该次运行的 SFT 验证分片 |
-| `wheels/`、`bin/setup.sh`、`bin/chronos` | 把内核和扩展两个 wheel 离线安装到 `.venv`；启动脚本设置 `CHRONOS_BUNDLE`，并把 `bundle …` 转给 `chronos-translate` |
-| `recipes/source_run_spec.json`、`recommended.json` | 产生这份权重的配方；微调档位 `smoke` / `quick` / `standard` / `full` |
-| `bundle.json` | 来源信息（步数、token 数、valid CE）、实测质量探针、每个文件的 SHA-256 |
-
-接收方的用法：
-
-```bash
-bin/setup.sh                                              # 建 venv、装 wheels、运行 `bundle check`
-bin/chronos bundle translate "The weather is very nice today."
-bin/chronos bundle translate --backend mlx --speculative --file examples/translate_en.txt
-bin/chronos bundle eval --split devtest -n 200
-bin/chronos bundle prepare-data my_pairs.jsonl --out work/data      # 会从你的数据中剔除 FLORES+
-bin/chronos bundle finetune --data work/data --run-dir work/ft --tier quick --mode sft
-bin/chronos bundle serve -- --port 11435
-```
-
-`--speculative` 要求 `--repetition-penalty 1.0`（此时的默认值），因为惩罚会逐位置改变目标分布。`finetune` 写出一份 `init_checkpoint` 指向包内权重的 `chronos-translate run` spec；运行目录里已有训练时会拒绝写入。用许可未知的数据（WMT19、OPUS-100）训练的模型包，`NOTICE.md` 中写明权重仅供研究使用。
-
 ### 部署导出
 
 ```bash
@@ -806,11 +753,15 @@ tuner.start(
         tune_lambda_lookahead=True, tune_lookahead_steps=True,
         tune_hidden_size=True, tune_num_experts=True,
         tune_num_shared_experts=True, tune_kv_latent_dim=True,
+        tune_recurrent_order=True,      # 用循环模型做探测（probe_recurrent_depth，默认 4）
+        tune_recurrent_dt_init=True,
     ),
     n_trials=20,
 )
 ```
 
+探测用的是 `dataset_path` 中的真实 token（最多 40 万个，用模型自己的分词器编码，没有则用内置 32k 分词器）；读不出数据时退回随机 id，并写一行日志。两个循环参数只能做粗筛：对 4 层模型训练几步，分辨不出差别小于运行间噪声的步长。`recurrent_later_groups` 不参与搜索，因为它是用质量换成本，只看损失的搜索永远会选完整的块。
+
 ---
 
 ## 项目结构

diff --git a/chronos/__init__.py b/chronos/__init__.py
@@ -1,5 +1,5 @@
 # Project Chronos: On-device low-latency lookahead dual-layer MoE inference
-__version__ = "0.3.2a3"
+__version__ = "0.3.3a0"
 
 # Register with HuggingFace AutoXXX factories so
 # AutoModelForCausalLM.from_pretrained("path/to/chronos") works out of the box.