Publish complete APUS-OpenJev-v1 models and Technical Report v1.1

.gitattributes

@@ -33,3 +33,7 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+4B-5949/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+9B-3000/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+9B-5949/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+TECHNICAL_REPORT.pdf filter=lfs diff=lfs merge=lfs -text

4B-5949/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2026 Alibaba Cloud
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

4B-5949/README.md

@@ -0,0 +1,33 @@
+# APUS-OpenJev-v1 · 4B
+
+A decision model for browser agents and business workflows. This directory contains standalone BF16 weights and a runtime with selectable `effort="low"` and `effort="high"` compute budgets.
+
+[Model family](../README.md) · [Architecture](../ARCHITECTURE.md) · [Runtime guide](RUNTIME.md)
+
+## Quick start
+
+Use a CUDA-capable PyTorch environment.
+
+```bash
+python -m pip install huggingface_hub
+hf auth login
+hf download apus-ailab/APUS-OpenJev-v1 \
+  --include "4B-5949/*" --local-dir ./APUS-OpenJev-v1
+cd ./APUS-OpenJev-v1/4B-5949
+python -m pip install -r requirements.txt
+python examples.py . --device cuda:0 --effort high
+```
+
+The included runtime provides compute-budget selection. Use `high` for text generation.
+
+## Evaluation
+
+With the full compute budget, this merged model scores **66/80 (82.50%)** on the [Frozen80 development panel](https://huggingface.co/datasets/gump2049/xDAN-openJet-Eval-Frozen80-20260921): browser action selection, principle-based judgment, evidence-based questions, natural language inference, and attribute decisions.
+
+This reused development panel is an engineering reference, not an independent benchmark. BF16 merging changes some candidate probabilities; decision thresholds require revalidation. See [evaluation results](merged-evaluation.json) and [runtime checks](evaluation/runtime-smoke.json) for details.
+
+## Provenance
+
+We thank the Qwen team for the [Qwen3.5-4B](https://huggingface.co/Qwen/Qwen3.5-4B) base model. Training details and source records are in [training.md](training.md); artifact hashes are in [release-manifest.json](release-manifest.json). Consult [LICENSE](LICENSE) and the [base-model](provenance/BASE-LICENSE.txt) and [source-project](provenance/SOURCE-PROJECT-LICENSE.txt) notices.
+
+**Authors:** gumpcheng ([https://huggingface.co/xDAN2099](https://huggingface.co/xDAN2099)), zhangxu, [APUS AI-LAB](https://github.com/APUS-AI-Lab)

4B-5949/RUNTIME.md

@@ -0,0 +1,55 @@
+# xDAN-openJet merged reference runtime
+
+本目录是可随 HF merged 仓库发布的完整 Python 源码闭包，不需要安装 ms-swift、PEFT 或原项目。发布选择为 **4B checkpoint-5949**、**9B checkpoint-3000** 与 **9B checkpoint-5949**；各目录必须保留自己的 `depth_config.json`、完整 Qwen config、tokenizer、chat template、generation config 和 merged safetensors。不得将不同 checkpoint 的权重或 shallow/full 结果拼在一起。
+
+## 运行
+
+使用 CUDA 对应 PyTorch 2.8.0 构建，安装 `requirements.txt`。运行依赖 Transformer 私有模型层接口，所以严格要求 `transformers==5.16.1`；版本升级需重做层级及数值验证。这是原生 PyTorch 单卡单请求参考实现，不是 vLLM 服务。
+
+先把发布仓库的**固定 commit**完整下载到本机目录。仓库根目录有本目录中的 `openjet_runtime/` 与 `examples.py` 时：
+
+```bash
+python -m pip install -r requirements.txt
+python examples.py ./model-snapshot --device cuda:0 --effort both
+python examples.py ./model-snapshot --device cuda:0 --effort high --text
+```
+
+若代码与权重同在下载快照根目录，进入快照后把 `./model-snapshot` 改为 `.`。
+
+```python
+from openjet_runtime import OpenJet
+from examples import decision_examples
+
+model = OpenJet.from_pretrained("./model-snapshot")
+two_candidates, sixteen_candidates = decision_examples()
+print(model.decide(two_candidates, effort="low"))
+print(model.decide(sixteen_candidates, effort="high"))
+print(model.generate_text("Return only the text: red shoes", effort="high", max_new_tokens=32))
+```
+
+`examples.py` 中两候选工作流、16 候选浏览器和 TYPE 是接口演示，不是声称模型已通过的 benchmark。需要针对业务设计 prompt 和验证答案。
+
+## 接口和执行语义
+
+- `decide(request, effort)` 输入字段与原 `jev.dynamic.prompt.v2` 相同：`id/group_id/state/instructions/primitive/criteria`；每个候选有非空 `id` 和 `description`，2–16 个，ID 不重复。标签为 A–P，编译器验证每个标签在真实回答边界恰为一个 token。没有 gold 输入需求。
+- `primitive="choice"` 返回 `choice` 与按输入顺序映射的 `probabilities`；`noul/score_level` 使用 `contracts.py` 中固定 Yes/No 候选，返回 `yes_probability`。`score_level` 是单个命题的判断，不能当作完整序数 Score API。
+- `effort="low"` 执行 `depth_config.exit_depth`（这两项发布预期16），共享原 LM final norm 与候选行投影；`high` 执行 `full_depth`（预期32），保持原评测中的标准模型前向+完整 LM head 路径。读取 config，不凭参数规模推测层数。
+- 所有输入采用 tokenizer 自带 chat template、`enable_thinking=False`，超过8192 token直接报错。不会默默截断 state、instructions 或候选。
+- `probabilities` 是在当前候选集上的相对 softmax，**未经概率校准**；合并不自动带来可信置信度或校准保证。
+- `generate_text` 为 TYPE 文本保留的贪心参考路径；每个 token 重新计算前缀，方便与原评测逐 token 核对，但不适合宣传 tokens/s。达到上限明确返回 `finish_reason="length"`。
+- `both` 示例分别调用两个 effort；没有自动路由、不承诺共享两次调用的前缀缓存。本次便携发布不包含 KV 广播引擎、vLLM 插件、TypeSafe HTTP server 或多模态输入能力。
+
+## 合并验收（GPU，不能用静态测试代替）
+
+1. 固定同一 base revision、adapter SHA、tokenizer、chat template、dtype、attention backend 和 Transformers 版本，记录 merge 前后权重身份。保留 adapter 原文件。
+2. 同进程/新进程分别加载 base+adapter 和 merged；比较固定2候选、16候选、长输入、80题面板两 effort 的 token IDs、候选排序、logits、probabilities 和最终ID。保存逐题差异与最大绝对差，不能只比 aggregate accuracy。
+3. BF16 merge 会舍入：不能预先声明 bitwise一致或把漂移简单解释为无害；应报告数值误差和所有预测翻转。若超过事先制定的容忍值，停止发布数值等价结论，考虑 FP32 merge/存储再独立评测。
+4. fresh reload 验证 `depth_config` 与实际层数、模块边界一致。用层 forward hooks 检查 low只执行浅层、high执行全层；hooks会触发候选头保守fallback，不拿该测量做性能报告。
+5. TYPE短样本比较 token序列和EOS；分别测试空输入、非法候选/重复ID、超长输入明确失败。
+6. 在干净环境固定 HF revision 下载，运行本目录例子和同一小面板。记录显存、依赖、GPU型号以及权重checksum。成功加载只是第一关，不能当作质量或吞吐验收。
+
+原始数值等价门结果：`False`，保留原结果；决策完全一致门：`True`，一致 `160/160`。独立运行时 GPU 验收状态：`passed`。若数值门失败，此 BF16 包作为独立重评版本，禁止直接迁移概率/拒答/路由阈值；见 `merged-evaluation.json`。
+
+## 源码来历
+
+`contracts.py`、`candidate_projection.py` 与 `early_exit.py` 从已有本地实现原样提取（最后一项仅调整相对 import）；`source-provenance.json` 记录源路径与两端 SHA256。`runtime.py` 是最小加载及接口层；high 与 TYPE 分别对应原 `HFDecisionEngine._forward_batch` 和 `package_eval.prefix_next_token` 的执行语义。采用现有模型类：`qwen3_5` → `Qwen3_5ForConditionalGeneration`；`qwen3_5_text` → `AutoModelForCausalLM`。没有新增学习参数。

4B-5949/chat_template.jinja

@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}

4B-5949/config.json

@@ -0,0 +1,109 @@
+{
+  "architectures": [
+    "Qwen3_5ForConditionalGeneration"
+  ],
+  "dtype": "bfloat16",
+  "image_token_id": 248056,
+  "model_type": "qwen3_5",
+  "text_config": {
+    "attention_bias": false,
+    "attention_dropout": 0.0,
+    "attn_output_gate": true,
+    "bos_token_id": null,
+    "dtype": "bfloat16",
+    "eos_token_id": 248044,
+    "full_attention_interval": 4,
+    "head_dim": 256,
+    "hidden_act": "silu",
+    "hidden_size": 2560,
+    "initializer_range": 0.02,
+    "intermediate_size": 9216,
+    "layer_types": [
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention"
+    ],
+    "linear_conv_kernel_dim": 4,
+    "linear_key_head_dim": 128,
+    "linear_num_key_heads": 16,
+    "linear_num_value_heads": 32,
+    "linear_value_head_dim": 128,
+    "mamba_ssm_dtype": "float32",
+    "max_position_embeddings": 262144,
+    "mlp_only_layers": [],
+    "model_type": "qwen3_5_text",
+    "mtp_num_hidden_layers": 1,
+    "mtp_use_dedicated_embeddings": false,
+    "num_attention_heads": 16,
+    "num_hidden_layers": 32,
+    "num_key_value_heads": 4,
+    "pad_token_id": null,
+    "partial_rotary_factor": 0.25,
+    "rms_norm_eps": 1e-06,
+    "rope_parameters": {
+      "mrope_interleaved": true,
+      "mrope_section": [
+        11,
+        11,
+        10
+      ],
+      "partial_rotary_factor": 0.25,
+      "rope_theta": 10000000,
+      "rope_type": "default"
+    },
+    "tie_word_embeddings": true,
+    "use_cache": true,
+    "vocab_size": 248320
+  },
+  "tie_word_embeddings": true,
+  "transformers_version": "5.16.1",
+  "video_token_id": 248057,
+  "vision_config": {
+    "deepstack_visual_indexes": [],
+    "depth": 24,
+    "dtype": "bfloat16",
+    "hidden_act": "gelu_pytorch_tanh",
+    "hidden_size": 1024,
+    "in_channels": 3,
+    "initializer_range": 0.02,
+    "intermediate_size": 4096,
+    "model_type": "qwen3_5_vision",
+    "num_heads": 16,
+    "num_position_embeddings": 2304,
+    "out_hidden_size": 2560,
+    "patch_size": 16,
+    "spatial_merge_size": 2,
+    "temporal_patch_size": 2
+  },
+  "vision_end_token_id": 248054,
+  "vision_start_token_id": 248053
+}

4B-5949/decision-release.json

@@ -0,0 +1,35 @@
+{
+  "passed": true,
+  "scope": "Independent BF16 merged variant, frozen80 decision parity; NOT a probability-equivalent replacement",
+  "post_observation_scope_amendment": true,
+  "original_numerical_gate_passed": false,
+  "original_gate": "identical160_argmax_and_max_probability_delta_le_0.05",
+  "original_comparison_sha256": "97cc94ddebb4f321d3d4ed29c89a0d8b1b3d6570e48bff427c635ba82f67862e",
+  "decision_agreement": 160,
+  "total_decisions": 160,
+  "independent_questions": 80,
+  "probability_calibration_transfer_validated": false,
+  "automatic_routing_validated": false,
+  "required_followup": "Recalibrate all probability, rejection and routing thresholds; independently evaluate new tasks",
+  "depths": {
+    "16": {
+      "before_correct": 61,
+      "after_correct": 61,
+      "changed_decisions": [],
+      "max_probability_abs_difference": 0.09226870536804199,
+      "mean_probability_abs_difference": 0.001908167676742778,
+      "max_logit_abs_difference": 0.5078125,
+      "mean_logit_abs_difference": 0.026137218475341797
+    },
+    "32": {
+      "before_correct": 66,
+      "after_correct": 66,
+      "changed_decisions": [],
+      "max_probability_abs_difference": 0.06145721673965454,
+      "mean_probability_abs_difference": 0.0017077110185891797,
+      "max_logit_abs_difference": 0.875,
+      "mean_logit_abs_difference": 0.03171875
+    }
+  },
+  "publication_requires": "Further portable-runtime160, weight arithmetic, full file hashes and fresh HF reload gates"
+}

4B-5949/depth_config.json

@@ -0,0 +1,10 @@
+{
+  "prompt_version": "jev.dynamic.prompt.v2",
+  "exit_depth": 16,
+  "full_depth": 32,
+  "model_series": "xDAN-openJet",
+  "checkpoint_step": 5949,
+  "source_depth_config_sha256": "6856cf257aa6eeb24dda20702ace04696b5d3db19d647705dedd00cca6f756e6",
+  "training_mode": "two_exit",
+  "automatic_routing_validated": false
+}

4B-5949/evaluation/runtime-smoke.json

@@ -0,0 +1,187 @@
+{
+  "status": "passed",
+  "runtime_sha256": {
+    "openjet_runtime/__init__.py": "df6eb864cf6d0c2f512fb00f17eeaa11fd790afca33a0f7b2f609aeb1f2b3944",
+    "openjet_runtime/runtime.py": "6e7b0b131cb14ab0d25cc8fd6c7fc41738b799cfe6de1ccdbae2d09d7c64313c",
+    "openjet_runtime/early_exit.py": "89b7751a927a6d1348a454fb8ee986395423ad5686d855fb3ccab255c95d2566",
+    "openjet_runtime/contracts.py": "d8e8e5270ecd6dab917d886dda2d684c24696b811faa968bb3c399ceef5e356a",
+    "openjet_runtime/candidate_projection.py": "84dc4746b5fb06ac6a9024dde3ba8414d901acf2a62d010b0d66f26acfaf74a6"
+  },
+  "decisions": 160,
+  "max_probability_delta": 0.0,
+  "no_jev_import": true,
+  "long_input_rejected": true,
+  "invalid_effort_rejected": true,
+  "synthetic_examples": [
+    {
+      "id": "example-binary",
+      "type": "choice",
+      "probabilities": {
+        "close": 0.9961305856704712,
+        "refund": 0.0038693908136337996
+      },
+      "choice": "close",
+      "effort": "low",
+      "executed_layers": 16,
+      "prompt_tokens": 103,
+      "logits": [
+        5.09375,
+        -0.45703125
+      ],
+      "projection": "candidate_rows",
+      "calibrated": false
+    },
+    {
+      "id": "example-binary",
+      "type": "choice",
+      "probabilities": {
+        "close": 0.9996199607849121,
+        "refund": 0.0003799845289904624
+      },
+      "choice": "close",
+      "effort": "high",
+      "executed_layers": 32,
+      "prompt_tokens": 103,
+      "logits": [
+        20.375,
+        12.5
+      ],
+      "projection": "full_head",
+      "calibrated": false
+    },
+    {
+      "id": "example-browser",
+      "type": "choice",
+      "probabilities": {
+        "click-1": 0.06669197976589203,
+        "click-2": 0.016471756622195244,
+        "click-3": 0.03487071022391319,
+        "click-4": 0.06072373315691948,
+        "click-5": 0.025511953979730606,
+        "click-6": 0.03125790134072304,
+        "click-7": 0.0377042256295681,
+        "click-8": 0.05234713479876518,
+        "click-9": 0.07498379796743393,
+        "click-10": 0.027584997937083244,
+        "click-11": 0.04879289120435715,
+        "click-12": 0.042062100023031235,
+        "click-13": 0.14767390489578247,
+        "click-14": 0.10910079628229141,
+        "click-15": 0.08630583435297012,
+        "click-16": 0.13791632652282715
+      },
+      "choice": "click-13",
+      "effort": "low",
+      "executed_layers": 16,
+      "prompt_tokens": 383,
+      "logits": [
+        -1.1796875,
+        -2.578125,
+        -1.828125,
+        -1.2734375,
+        -2.140625,
+        -1.9375,
+        -1.75,
+        -1.421875,
+        -1.0625,
+        -2.0625,
+        -1.4921875,
+        -1.640625,
+        -0.384765625,
+        -0.6875,
+        -0.921875,
+        -0.453125
+      ],
+      "projection": "candidate_rows",
+      "calibrated": false
+    },
+    {
+      "id": "example-browser",
+      "type": "choice",
+      "probabilities": {
+        "click-1": 0.00045352030429057777,
+        "click-2": 0.00040023025940172374,
+        "click-3": 0.00017760110495146364,
+        "click-4": 0.0002142278099199757,
+        "click-5": 0.0002584080502856523,
+        "click-6": 0.00024275195028167218,
+        "click-7": 0.0003532019618432969,
+        "click-8": 0.0004827698867302388,
+        "click-9": 0.0013123045209795237,
+        "click-10": 0.0004827698867302388,
+        "click-11": 0.0035672136582434177,
+        "click-12": 0.9890894293785095,
+        "click-13": 0.0012327961158007383,
+        "click-14": 0.0007959529175423086,
+        "click-15": 0.0001890553830889985,
+        "click-16": 0.0007477285689674318
+      },
+      "choice": "click-12",
+      "effort": "high",
+      "executed_layers": 32,
+      "prompt_tokens": 383,
+      "logits": [
+        12.5625,
+        12.4375,
+        11.625,
+        11.8125,
+        12.0,
+        11.9375,
+        12.3125,
+        12.625,
+        13.625,
+        12.625,
+        14.625,
+        20.25,
+        13.5625,
+        13.125,
+        11.6875,
+        13.0625
+      ],
+      "projection": "full_head",
+      "calibrated": false
+    }
+  ],
+  "text_smoke": {
+    "low": {
+      "text": "\u4f60\u597d / / / / ...etc\u4e4e--SsN [sS",
+      "token_ids": [
+        109266,
+        593,
+        593,
+        593,
+        593,
+        2423,
+        11763,
+        97091,
+        12,
+        12,
+        50,
+        82,
+        45,
+        498,
+        82,
+        50
+      ],
+      "effort": "low",
+      "executed_layers_per_token": 16,
+      "finish_reason": "length",
+      "prompt_tokens": 28
+    },
+    "high": {
+      "text": "red shoes\n",
+      "token_ids": [
+        1114,
+        14850,
+        248046,
+        198,
+        248044
+      ],
+      "effort": "high",
+      "executed_layers_per_token": 32,
+      "finish_reason": "eos",
+      "prompt_tokens": 28
+    }
+  },
+  "text_scope": "Execution smoke only; not TYPE accuracy or speed validation"
+}

4B-5949/evaluation/weight-arithmetic.json

@@ -0,0 +1,908 @@
+{
+  "status": "passed",
+  "base_mtp_keys_not_loaded_by_transformers": [
+    "mtp.fc.weight",
+    "mtp.layers.0.input_layernorm.weight",
+    "mtp.layers.0.mlp.down_proj.weight",
+    "mtp.layers.0.mlp.gate_proj.weight",
+    "mtp.layers.0.mlp.up_proj.weight",
+    "mtp.layers.0.post_attention_layernorm.weight",
+    "mtp.layers.0.self_attn.k_norm.weight",
+    "mtp.layers.0.self_attn.k_proj.weight",
+    "mtp.layers.0.self_attn.o_proj.weight",
+    "mtp.layers.0.self_attn.q_norm.weight",
+    "mtp.layers.0.self_attn.q_proj.weight",
+    "mtp.layers.0.self_attn.v_proj.weight",
+    "mtp.norm.weight",
+    "mtp.pre_fc_norm_embedding.weight",
+    "mtp.pre_fc_norm_hidden.weight"
+  ],
+  "size": "4B",
+  "targeted_weight_matrices": 176,
+  "adapter_tensors": 352,
+  "all_targeted_weights_exact": true,
+  "scope": "FP32 LoRA arithmetic followed by BF16 rounding; not forward/probability equivalence",
+  "adapter_sha256": "71154f60ec72c55d2c6c6147b9cdda9cc9a52d46297f3074dded7cdc9f5bc344",
+  "checks": [
+    {
+      "weight": "model.language_model.layers.0.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.0.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.0.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.0.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.0.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.1.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.1.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.1.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.1.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.1.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.10.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.10.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.10.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.10.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.10.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.11.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.11.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.11.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.11.self_attn.k_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.11.self_attn.o_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.11.self_attn.q_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.11.self_attn.v_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.12.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.12.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.12.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.12.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.12.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.13.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.13.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.13.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.13.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.13.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.14.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.14.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.14.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.14.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.14.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.15.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.15.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.15.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.15.self_attn.k_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.15.self_attn.o_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.15.self_attn.q_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.15.self_attn.v_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.16.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.16.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.16.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.16.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.16.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.17.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.17.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.17.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.17.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.17.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.18.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.18.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.18.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.18.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.18.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.19.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.19.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.19.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.19.self_attn.k_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.19.self_attn.o_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.19.self_attn.q_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.19.self_attn.v_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.2.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.2.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.2.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.2.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.2.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.20.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.20.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.20.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.20.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.20.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.21.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.21.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.21.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.21.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.21.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.22.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.22.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.22.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.22.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.22.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.23.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.23.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.23.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.23.self_attn.k_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.23.self_attn.o_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.23.self_attn.q_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.23.self_attn.v_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.24.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.24.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.24.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.24.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.24.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.25.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.25.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.25.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.25.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.25.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.26.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.26.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.26.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.26.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.26.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.27.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.27.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.27.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.27.self_attn.k_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.27.self_attn.o_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.27.self_attn.q_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.27.self_attn.v_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.28.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.28.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.28.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.28.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.28.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.29.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.29.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.29.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.29.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.29.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.3.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.3.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.3.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.3.self_attn.k_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.3.self_attn.o_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.3.self_attn.q_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.3.self_attn.v_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.30.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.30.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.30.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.30.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.30.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.31.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.31.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.31.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.31.self_attn.k_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.31.self_attn.o_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.31.self_attn.q_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.31.self_attn.v_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.4.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.4.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.4.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.4.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.4.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.5.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.5.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.5.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.5.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.5.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.6.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.6.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.6.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.6.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.6.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.7.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.7.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.7.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.7.self_attn.k_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.7.self_attn.o_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.7.self_attn.q_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.7.self_attn.v_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.8.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.8.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.8.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.8.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.8.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.9.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.9.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.9.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.9.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.9.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    }
+  ]
+}

4B-5949/examples.py

@@ -0,0 +1,61 @@
+"""Run against a local merged HF snapshot; no project-local dependencies."""
+
+import argparse
+import json
+
+from openjet_runtime import OpenJet
+
+
+def decision_examples():
+    binary = {
+        "id": "example-binary",
+        "group_id": "example-binary",
+        "primitive": "choice",
+        "state": "Order 731 has been delivered. The customer's message says thank you.",
+        "instructions": "Select the appropriate next workflow action.",
+        "criteria": [
+            {"id": "close", "description": "Close the resolved support ticket."},
+            {"id": "refund", "description": "Refund an undelivered order."},
+        ],
+    }
+    browser = {
+        "id": "example-browser",
+        "group_id": "example-browser",
+        "primitive": "choice",
+        "state": "A settings page has 16 visible buttons labeled Page 1 through Page 16.",
+        "instructions": "Navigate to Page 12 by choosing its matching button.",
+        "criteria": [
+            {"id": f"click-{i}", "description": f"Click the Page {i} button."}
+            for i in range(1, 17)
+        ],
+    }
+    return binary, browser
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("model", help="Local merged snapshot directory")
+    parser.add_argument("--device", default="cuda:0")
+    parser.add_argument("--dtype", choices=["float32", "bfloat16"], default="bfloat16")
+    parser.add_argument("--effort", choices=["low", "high", "both"], default="both")
+    parser.add_argument(
+        "--text", action="store_true", help="Also run slow TYPE reference"
+    )
+    args = parser.parse_args()
+    runtime = OpenJet.from_pretrained(args.model, args.device, args.dtype)
+    efforts = ("low", "high") if args.effort == "both" else (args.effort,)
+    for effort in efforts:
+        for request in decision_examples():
+            result = runtime.decide(request, effort)
+            print(json.dumps({"example": request["id"], **result}, ensure_ascii=False))
+        if args.text:
+            result = runtime.generate_text(
+                "Return only the literal text to type into a search box for 'red shoes'.",
+                effort=effort,
+                max_new_tokens=32,
+            )
+            print(json.dumps({"example": "browser-type", **result}, ensure_ascii=False))
+
+
+if __name__ == "__main__":
+    main()

4B-5949/generation_config.json

@@ -0,0 +1,6 @@
+{
+  "_from_model_config": true,
+  "eos_token_id": 248044,
+  "transformers_version": "5.16.1",
+  "use_cache": true
+}

4B-5949/merge-provenance.json

@@ -0,0 +1,17 @@
+{
+  "base_id": "Qwen/Qwen3.5-4B",
+  "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a",
+  "checkpoint_step": 5949,
+  "adapter_sha256": "71154f60ec72c55d2c6c6147b9cdda9cc9a52d46297f3074dded7cdc9f5bc344",
+  "adapter_config_sha256": "c6caee3818ca1f6c8539e47fac9b9fa818d28b61a4bbb05f6f8444c0dd639e45",
+  "official80_sha256": "b3374e82f0e605762d40ab6455449c9cb2d315804a175d1cda985cba9beded35",
+  "software": {
+    "torch": "2.8.0+cu128",
+    "transformers": "5.16.1",
+    "peft": "0.20.0"
+  },
+  "merge_arithmetic": "float32 CPU safe_merge then bfloat16 storage",
+  "inference_dtype": "bfloat16",
+  "attention": "sdpa",
+  "gpu": "NVIDIA RTX PRO 6000 Blackwell Server Edition"
+}

4B-5949/merged-evaluation.json

@@ -0,0 +1,83 @@
+{
+  "scope": "80 fixed requests, both explicit efforts; merged-model comparison, not independent generalization evidence",
+  "checkpoint_step": 5949,
+  "base_id": "Qwen/Qwen3.5-4B",
+  "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a",
+  "comparison": {
+    "passed": false,
+    "gate": "identical160_argmax_and_max_probability_delta_le_0.05",
+    "depths": {
+      "16": {
+        "before_correct": 61,
+        "after_correct": 61,
+        "changed_decisions": [],
+        "max_probability_abs_difference": 0.09226870536804199,
+        "mean_probability_abs_difference": 0.001908167676742778,
+        "max_logit_abs_difference": 0.5078125,
+        "mean_logit_abs_difference": 0.026137218475341797
+      },
+      "32": {
+        "before_correct": 66,
+        "after_correct": 66,
+        "changed_decisions": [],
+        "max_probability_abs_difference": 0.06145721673965454,
+        "mean_probability_abs_difference": 0.0017077110185891797,
+        "max_logit_abs_difference": 0.875,
+        "mean_logit_abs_difference": 0.03171875
+      }
+    },
+    "before_sha256": "7be98e54b67be84ae97a4152adeddd47a351e224cbe9345bae8ca8f44c265964",
+    "after_sha256": "ac22ede19c6ade99388038292644df50e9891dff162acdb10029bb3bd343537e",
+    "verified_at_unix": 1789983995.6865954
+  },
+  "runtime_validation": {
+    "status": "passed",
+    "decisions": 160,
+    "max_probability_delta": 0.0,
+    "no_jev_import": true,
+    "long_input_rejected": true,
+    "invalid_effort_rejected": true,
+    "text_scope": "Execution smoke only; not TYPE accuracy or speed validation",
+    "source_sha256": "60a80902675971ff6a054377852b936eec90f49f466eb7b8b077eee7b979f230"
+  },
+  "decision_release": {
+    "passed": true,
+    "scope": "Independent BF16 merged variant, frozen80 decision parity; NOT a probability-equivalent replacement",
+    "post_observation_scope_amendment": true,
+    "original_numerical_gate_passed": false,
+    "original_gate": "identical160_argmax_and_max_probability_delta_le_0.05",
+    "original_comparison_sha256": "97cc94ddebb4f321d3d4ed29c89a0d8b1b3d6570e48bff427c635ba82f67862e",
+    "decision_agreement": 160,
+    "total_decisions": 160,
+    "independent_questions": 80,
+    "probability_calibration_transfer_validated": false,
+    "automatic_routing_validated": false,
+    "required_followup": "Recalibrate all probability, rejection and routing thresholds; independently evaluate new tasks",
+    "depths": {
+      "16": {
+        "before_correct": 61,
+        "after_correct": 61,
+        "changed_decisions": [],
+        "max_probability_abs_difference": 0.09226870536804199,
+        "mean_probability_abs_difference": 0.001908167676742778,
+        "max_logit_abs_difference": 0.5078125,
+        "mean_logit_abs_difference": 0.026137218475341797
+      },
+      "32": {
+        "before_correct": 66,
+        "after_correct": 66,
+        "changed_decisions": [],
+        "max_probability_abs_difference": 0.06145721673965454,
+        "mean_probability_abs_difference": 0.0017077110185891797,
+        "max_logit_abs_difference": 0.875,
+        "mean_logit_abs_difference": 0.03171875
+      }
+    },
+    "publication_requires": "Further portable-runtime160, weight arithmetic, full file hashes and fresh HF reload gates"
+  },
+  "reviewed_variant": {},
+  "decision_parity_passed": true,
+  "decision_agreement": 160,
+  "weight_arithmetic_status": "passed",
+  "raw_inputs_included": false
+}

4B-5949/model-00001-of-00003.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c04bef62040612a3376c144014c194687cdc19b18c3a807ee1136b4a713cbb0a
+size 3991298872

4B-5949/model-00002-of-00003.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7891d5b76b686893680e9cc074c2e17a788ff0cb03f64cc2ae4150bd804299a2
+size 3979833152

4B-5949/model-00003-of-00003.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a88eecd668f83cad773bd67c9c7e6e466c1746d489c55d906b48398a6679db65
+size 1107487880

4B-5949/model.safetensors.index.json

@@ -0,0 +1,731 @@
+{
+  "metadata": {
+    "total_parameters": 4539265536,
+    "total_size": 9078531072
+  },
+  "weight_map": {
+    "model.language_model.embed_tokens.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.0.input_layernorm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.0.linear_attn.A_log": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.0.linear_attn.conv1d.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.0.linear_attn.dt_bias": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.0.linear_attn.in_proj_a.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.0.linear_attn.in_proj_b.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.0.linear_attn.in_proj_qkv.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.0.linear_attn.in_proj_z.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.0.linear_attn.norm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.0.linear_attn.out_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.0.mlp.down_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.0.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.0.mlp.up_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.0.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.1.input_layernorm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.1.linear_attn.A_log": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.1.linear_attn.conv1d.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.1.linear_attn.dt_bias": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.1.linear_attn.in_proj_a.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.1.linear_attn.in_proj_b.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.1.linear_attn.in_proj_qkv.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.1.linear_attn.in_proj_z.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.1.linear_attn.norm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.1.linear_attn.out_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.1.mlp.down_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.1.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.1.mlp.up_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.1.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.10.input_layernorm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.10.linear_attn.A_log": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.10.linear_attn.conv1d.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.10.linear_attn.dt_bias": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.10.linear_attn.in_proj_a.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.10.linear_attn.in_proj_b.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.10.linear_attn.in_proj_qkv.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.10.linear_attn.in_proj_z.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.10.linear_attn.norm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.10.linear_attn.out_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.10.mlp.down_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.10.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.10.mlp.up_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.10.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.11.input_layernorm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.11.mlp.down_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.11.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.11.mlp.up_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.11.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.11.self_attn.k_norm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.11.self_attn.k_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.11.self_attn.o_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.11.self_attn.q_norm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.11.self_attn.q_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.11.self_attn.v_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.12.input_layernorm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.12.linear_attn.A_log": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.12.linear_attn.conv1d.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.12.linear_attn.dt_bias": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.12.linear_attn.in_proj_a.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.12.linear_attn.in_proj_b.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.12.linear_attn.in_proj_qkv.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.12.linear_attn.in_proj_z.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.12.linear_attn.norm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.12.linear_attn.out_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.12.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.12.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.12.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.12.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.13.input_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.13.linear_attn.A_log": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.13.linear_attn.conv1d.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.13.linear_attn.dt_bias": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.13.linear_attn.in_proj_a.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.13.linear_attn.in_proj_b.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.13.linear_attn.in_proj_qkv.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.13.linear_attn.in_proj_z.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.13.linear_attn.norm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.13.linear_attn.out_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.13.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.13.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.13.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.13.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.14.input_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.14.linear_attn.A_log": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.14.linear_attn.conv1d.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.14.linear_attn.dt_bias": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.14.linear_attn.in_proj_a.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.14.linear_attn.in_proj_b.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.14.linear_attn.in_proj_qkv.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.14.linear_attn.in_proj_z.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.14.linear_attn.norm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.14.linear_attn.out_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.14.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.14.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.14.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.14.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.15.input_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.15.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.15.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.15.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.15.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.15.self_attn.k_norm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.15.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.15.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.15.self_attn.q_norm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.15.self_attn.q_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.15.self_attn.v_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.16.input_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.16.linear_attn.A_log": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.16.linear_attn.conv1d.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.16.linear_attn.dt_bias": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.16.linear_attn.in_proj_a.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.16.linear_attn.in_proj_b.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.16.linear_attn.in_proj_qkv.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.16.linear_attn.in_proj_z.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.16.linear_attn.norm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.16.linear_attn.out_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.16.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.16.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.16.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.16.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.17.input_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.17.linear_attn.A_log": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.17.linear_attn.conv1d.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.17.linear_attn.dt_bias": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.17.linear_attn.in_proj_a.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.17.linear_attn.in_proj_b.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.17.linear_attn.in_proj_qkv.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.17.linear_attn.in_proj_z.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.17.linear_attn.norm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.17.linear_attn.out_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.17.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.17.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.17.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.17.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.18.input_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.18.linear_attn.A_log": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.18.linear_attn.conv1d.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.18.linear_attn.dt_bias": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.18.linear_attn.in_proj_a.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.18.linear_attn.in_proj_b.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.18.linear_attn.in_proj_qkv.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.18.linear_attn.in_proj_z.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.18.linear_attn.norm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.18.linear_attn.out_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.18.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.18.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.18.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.18.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.19.input_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.19.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.19.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.19.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.19.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.19.self_attn.k_norm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.19.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.19.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.19.self_attn.q_norm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.19.self_attn.q_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.19.self_attn.v_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.2.input_layernorm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.2.linear_attn.A_log": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.2.linear_attn.conv1d.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.2.linear_attn.dt_bias": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.2.linear_attn.in_proj_a.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.2.linear_attn.in_proj_b.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.2.linear_attn.in_proj_qkv.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.2.linear_attn.in_proj_z.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.2.linear_attn.norm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.2.linear_attn.out_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.2.mlp.down_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.2.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.2.mlp.up_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.2.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.20.input_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.20.linear_attn.A_log": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.20.linear_attn.conv1d.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.20.linear_attn.dt_bias": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.20.linear_attn.in_proj_a.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.20.linear_attn.in_proj_b.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.20.linear_attn.in_proj_qkv.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.20.linear_attn.in_proj_z.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.20.linear_attn.norm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.20.linear_attn.out_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.20.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.20.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.20.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.20.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.21.input_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.21.linear_attn.A_log": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.21.linear_attn.conv1d.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.21.linear_attn.dt_bias": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.21.linear_attn.in_proj_a.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.21.linear_attn.in_proj_b.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.21.linear_attn.in_proj_qkv.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.21.linear_attn.in_proj_z.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.21.linear_attn.norm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.21.linear_attn.out_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.21.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.21.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.21.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.21.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.22.input_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.22.linear_attn.A_log": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.22.linear_attn.conv1d.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.22.linear_attn.dt_bias": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.22.linear_attn.in_proj_a.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.22.linear_attn.in_proj_b.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.22.linear_attn.in_proj_qkv.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.22.linear_attn.in_proj_z.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.22.linear_attn.norm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.22.linear_attn.out_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.22.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.22.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.22.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.22.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.23.input_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.23.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.23.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.23.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.23.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.23.self_attn.k_norm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.23.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.23.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.23.self_attn.q_norm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.23.self_attn.q_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.23.self_attn.v_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.24.input_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.24.linear_attn.A_log": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.24.linear_attn.conv1d.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.24.linear_attn.dt_bias": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.24.linear_attn.in_proj_a.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.24.linear_attn.in_proj_b.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.24.linear_attn.in_proj_qkv.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.24.linear_attn.in_proj_z.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.24.linear_attn.norm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.24.linear_attn.out_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.24.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.24.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.24.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.24.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.25.input_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.25.linear_attn.A_log": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.25.linear_attn.conv1d.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.25.linear_attn.dt_bias": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.25.linear_attn.in_proj_a.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.25.linear_attn.in_proj_b.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.25.linear_attn.in_proj_qkv.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.25.linear_attn.in_proj_z.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.25.linear_attn.norm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.25.linear_attn.out_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.25.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.25.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.25.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.25.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.26.input_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.26.linear_attn.A_log": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.26.linear_attn.conv1d.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.26.linear_attn.dt_bias": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.26.linear_attn.in_proj_a.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.26.linear_attn.in_proj_b.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.26.linear_attn.in_proj_qkv.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.26.linear_attn.in_proj_z.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.26.linear_attn.norm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.26.linear_attn.out_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.26.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.26.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.26.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.26.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.27.input_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.27.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.27.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.27.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.27.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.27.self_attn.k_norm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.27.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.27.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.27.self_attn.q_norm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.27.self_attn.q_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.27.self_attn.v_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.28.input_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.28.linear_attn.A_log": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.28.linear_attn.conv1d.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.28.linear_attn.dt_bias": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.28.linear_attn.in_proj_a.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.28.linear_attn.in_proj_b.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.28.linear_attn.in_proj_qkv.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.28.linear_attn.in_proj_z.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.28.linear_attn.norm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.28.linear_attn.out_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.28.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.28.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.28.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.28.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.29.input_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.29.linear_attn.A_log": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.29.linear_attn.conv1d.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.29.linear_attn.dt_bias": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.29.linear_attn.in_proj_a.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.29.linear_attn.in_proj_b.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.29.linear_attn.in_proj_qkv.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.29.linear_attn.in_proj_z.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.29.linear_attn.norm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.29.linear_attn.out_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.29.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.29.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.29.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.29.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.3.input_layernorm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.3.mlp.down_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.3.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.3.mlp.up_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.3.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.3.self_attn.k_norm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.3.self_attn.k_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.3.self_attn.o_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.3.self_attn.q_norm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.3.self_attn.q_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.3.self_attn.v_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.30.input_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.30.linear_attn.A_log": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.30.linear_attn.conv1d.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.30.linear_attn.dt_bias": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.30.linear_attn.in_proj_a.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.30.linear_attn.in_proj_b.weight": "model-00002-of-00003.safetensors",
+    "model.language_model.layers.30.linear_attn.in_proj_qkv.weight": "model-00003-of-00003.safetensors",
+    "model.language_model.layers.30.linear_attn.in_proj_z.weight": "model-00003-of-00003.safetensors",
+    "model.language_model.layers.30.linear_attn.norm.weight": "model-00003-of-00003.safetensors",
+    "model.language_model.layers.30.linear_attn.out_proj.weight": "model-00003-of-00003.safetensors",
+    "model.language_model.layers.30.mlp.down_proj.weight": "model-00003-of-00003.safetensors",
+    "model.language_model.layers.30.mlp.gate_proj.weight": "model-00003-of-00003.safetensors",
+    "model.language_model.layers.30.mlp.up_proj.weight": "model-00003-of-00003.safetensors",
+    "model.language_model.layers.30.post_attention_layernorm.weight": "model-00003-of-00003.safetensors",
+    "model.language_model.layers.31.input_layernorm.weight": "model-00003-of-00003.safetensors",
+    "model.language_model.layers.31.mlp.down_proj.weight": "model-00003-of-00003.safetensors",
+    "model.language_model.layers.31.mlp.gate_proj.weight": "model-00003-of-00003.safetensors",
+    "model.language_model.layers.31.mlp.up_proj.weight": "model-00003-of-00003.safetensors",
+    "model.language_model.layers.31.post_attention_layernorm.weight": "model-00003-of-00003.safetensors",
+    "model.language_model.layers.31.self_attn.k_norm.weight": "model-00003-of-00003.safetensors",
+    "model.language_model.layers.31.self_attn.k_proj.weight": "model-00003-of-00003.safetensors",
+    "model.language_model.layers.31.self_attn.o_proj.weight": "model-00003-of-00003.safetensors",
+    "model.language_model.layers.31.self_attn.q_norm.weight": "model-00003-of-00003.safetensors",
+    "model.language_model.layers.31.self_attn.q_proj.weight": "model-00003-of-00003.safetensors",
+    "model.language_model.layers.31.self_attn.v_proj.weight": "model-00003-of-00003.safetensors",
+    "model.language_model.layers.4.input_layernorm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.4.linear_attn.A_log": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.4.linear_attn.conv1d.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.4.linear_attn.dt_bias": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.4.linear_attn.in_proj_a.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.4.linear_attn.in_proj_b.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.4.linear_attn.in_proj_qkv.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.4.linear_attn.in_proj_z.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.4.linear_attn.norm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.4.linear_attn.out_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.4.mlp.down_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.4.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.4.mlp.up_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.4.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.5.input_layernorm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.5.linear_attn.A_log": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.5.linear_attn.conv1d.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.5.linear_attn.dt_bias": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.5.linear_attn.in_proj_a.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.5.linear_attn.in_proj_b.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.5.linear_attn.in_proj_qkv.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.5.linear_attn.in_proj_z.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.5.linear_attn.norm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.5.linear_attn.out_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.5.mlp.down_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.5.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.5.mlp.up_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.5.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.6.input_layernorm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.6.linear_attn.A_log": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.6.linear_attn.conv1d.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.6.linear_attn.dt_bias": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.6.linear_attn.in_proj_a.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.6.linear_attn.in_proj_b.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.6.linear_attn.in_proj_qkv.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.6.linear_attn.in_proj_z.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.6.linear_attn.norm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.6.linear_attn.out_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.6.mlp.down_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.6.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.6.mlp.up_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.6.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.7.input_layernorm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.7.mlp.down_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.7.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.7.mlp.up_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.7.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.7.self_attn.k_norm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.7.self_attn.k_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.7.self_attn.o_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.7.self_attn.q_norm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.7.self_attn.q_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.7.self_attn.v_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.8.input_layernorm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.8.linear_attn.A_log": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.8.linear_attn.conv1d.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.8.linear_attn.dt_bias": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.8.linear_attn.in_proj_a.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.8.linear_attn.in_proj_b.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.8.linear_attn.in_proj_qkv.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.8.linear_attn.in_proj_z.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.8.linear_attn.norm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.8.linear_attn.out_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.8.mlp.down_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.8.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.8.mlp.up_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.8.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.9.input_layernorm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.9.linear_attn.A_log": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.9.linear_attn.conv1d.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.9.linear_attn.dt_bias": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.9.linear_attn.in_proj_a.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.9.linear_attn.in_proj_b.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.9.linear_attn.in_proj_qkv.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.9.linear_attn.in_proj_z.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.9.linear_attn.norm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.9.linear_attn.out_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.9.mlp.down_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.9.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.9.mlp.up_proj.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.layers.9.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
+    "model.language_model.norm.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.0.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.0.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.0.attn.qkv.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.0.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.0.mlp.linear_fc1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.0.mlp.linear_fc1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.0.mlp.linear_fc2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.0.mlp.linear_fc2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.0.norm1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.0.norm1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.0.norm2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.0.norm2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.1.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.1.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.1.attn.qkv.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.1.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.1.mlp.linear_fc1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.1.mlp.linear_fc1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.1.mlp.linear_fc2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.1.mlp.linear_fc2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.1.norm1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.1.norm1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.1.norm2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.1.norm2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.10.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.10.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.10.attn.qkv.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.10.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.10.mlp.linear_fc1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.10.mlp.linear_fc1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.10.mlp.linear_fc2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.10.mlp.linear_fc2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.10.norm1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.10.norm1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.10.norm2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.10.norm2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.11.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.11.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.11.attn.qkv.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.11.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.11.mlp.linear_fc1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.11.mlp.linear_fc1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.11.mlp.linear_fc2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.11.mlp.linear_fc2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.11.norm1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.11.norm1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.11.norm2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.11.norm2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.12.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.12.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.12.attn.qkv.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.12.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.12.mlp.linear_fc1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.12.mlp.linear_fc1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.12.mlp.linear_fc2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.12.mlp.linear_fc2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.12.norm1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.12.norm1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.12.norm2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.12.norm2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.13.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.13.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.13.attn.qkv.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.13.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.13.mlp.linear_fc1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.13.mlp.linear_fc1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.13.mlp.linear_fc2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.13.mlp.linear_fc2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.13.norm1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.13.norm1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.13.norm2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.13.norm2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.14.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.14.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.14.attn.qkv.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.14.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.14.mlp.linear_fc1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.14.mlp.linear_fc1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.14.mlp.linear_fc2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.14.mlp.linear_fc2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.14.norm1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.14.norm1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.14.norm2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.14.norm2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.15.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.15.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.15.attn.qkv.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.15.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.15.mlp.linear_fc1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.15.mlp.linear_fc1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.15.mlp.linear_fc2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.15.mlp.linear_fc2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.15.norm1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.15.norm1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.15.norm2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.15.norm2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.16.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.16.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.16.attn.qkv.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.16.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.16.mlp.linear_fc1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.16.mlp.linear_fc1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.16.mlp.linear_fc2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.16.mlp.linear_fc2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.16.norm1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.16.norm1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.16.norm2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.16.norm2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.17.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.17.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.17.attn.qkv.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.17.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.17.mlp.linear_fc1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.17.mlp.linear_fc1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.17.mlp.linear_fc2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.17.mlp.linear_fc2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.17.norm1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.17.norm1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.17.norm2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.17.norm2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.18.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.18.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.18.attn.qkv.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.18.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.18.mlp.linear_fc1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.18.mlp.linear_fc1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.18.mlp.linear_fc2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.18.mlp.linear_fc2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.18.norm1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.18.norm1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.18.norm2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.18.norm2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.19.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.19.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.19.attn.qkv.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.19.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.19.mlp.linear_fc1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.19.mlp.linear_fc1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.19.mlp.linear_fc2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.19.mlp.linear_fc2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.19.norm1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.19.norm1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.19.norm2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.19.norm2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.2.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.2.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.2.attn.qkv.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.2.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.2.mlp.linear_fc1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.2.mlp.linear_fc1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.2.mlp.linear_fc2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.2.mlp.linear_fc2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.2.norm1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.2.norm1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.2.norm2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.2.norm2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.20.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.20.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.20.attn.qkv.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.20.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.20.mlp.linear_fc1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.20.mlp.linear_fc1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.20.mlp.linear_fc2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.20.mlp.linear_fc2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.20.norm1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.20.norm1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.20.norm2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.20.norm2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.21.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.21.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.21.attn.qkv.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.21.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.21.mlp.linear_fc1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.21.mlp.linear_fc1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.21.mlp.linear_fc2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.21.mlp.linear_fc2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.21.norm1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.21.norm1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.21.norm2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.21.norm2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.22.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.22.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.22.attn.qkv.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.22.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.22.mlp.linear_fc1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.22.mlp.linear_fc1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.22.mlp.linear_fc2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.22.mlp.linear_fc2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.22.norm1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.22.norm1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.22.norm2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.22.norm2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.23.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.23.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.23.attn.qkv.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.23.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.23.mlp.linear_fc1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.23.mlp.linear_fc1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.23.mlp.linear_fc2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.23.mlp.linear_fc2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.23.norm1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.23.norm1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.23.norm2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.23.norm2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.3.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.3.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.3.attn.qkv.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.3.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.3.mlp.linear_fc1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.3.mlp.linear_fc1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.3.mlp.linear_fc2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.3.mlp.linear_fc2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.3.norm1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.3.norm1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.3.norm2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.3.norm2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.4.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.4.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.4.attn.qkv.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.4.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.4.mlp.linear_fc1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.4.mlp.linear_fc1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.4.mlp.linear_fc2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.4.mlp.linear_fc2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.4.norm1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.4.norm1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.4.norm2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.4.norm2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.5.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.5.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.5.attn.qkv.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.5.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.5.mlp.linear_fc1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.5.mlp.linear_fc1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.5.mlp.linear_fc2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.5.mlp.linear_fc2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.5.norm1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.5.norm1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.5.norm2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.5.norm2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.6.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.6.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.6.attn.qkv.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.6.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.6.mlp.linear_fc1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.6.mlp.linear_fc1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.6.mlp.linear_fc2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.6.mlp.linear_fc2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.6.norm1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.6.norm1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.6.norm2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.6.norm2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.7.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.7.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.7.attn.qkv.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.7.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.7.mlp.linear_fc1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.7.mlp.linear_fc1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.7.mlp.linear_fc2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.7.mlp.linear_fc2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.7.norm1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.7.norm1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.7.norm2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.7.norm2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.8.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.8.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.8.attn.qkv.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.8.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.8.mlp.linear_fc1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.8.mlp.linear_fc1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.8.mlp.linear_fc2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.8.mlp.linear_fc2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.8.norm1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.8.norm1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.8.norm2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.8.norm2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.9.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.9.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.9.attn.qkv.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.9.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.9.mlp.linear_fc1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.9.mlp.linear_fc1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.9.mlp.linear_fc2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.9.mlp.linear_fc2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.9.norm1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.9.norm1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.9.norm2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.blocks.9.norm2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.merger.linear_fc1.bias": "model-00003-of-00003.safetensors",
+    "model.visual.merger.linear_fc1.weight": "model-00003-of-00003.safetensors",
+    "model.visual.merger.linear_fc2.bias": "model-00003-of-00003.safetensors",
+    "model.visual.merger.linear_fc2.weight": "model-00003-of-00003.safetensors",
+    "model.visual.merger.norm.bias": "model-00003-of-00003.safetensors",
+    "model.visual.merger.norm.weight": "model-00003-of-00003.safetensors",
+    "model.visual.patch_embed.proj.bias": "model-00003-of-00003.safetensors",
+    "model.visual.patch_embed.proj.weight": "model-00003-of-00003.safetensors",
+    "model.visual.pos_embed.weight": "model-00003-of-00003.safetensors"
+  }
+}

4B-5949/openjet_runtime/__init__.py

@@ -0,0 +1,3 @@
+from .runtime import OpenJet
+
+__all__ = ["OpenJet"]

4B-5949/openjet_runtime/candidate_projection.py

@@ -0,0 +1,108 @@
+"""Project selected native dense LM-head rows before matmul, preserving autograd.
+
+The caller must pass the actual output head, never an unwrapped adapter base_layer.
+Unsupported heads raise; there is deliberately no automatic full-head fallback.
+"""
+
+import torch
+from torch import nn
+from torch.nn import functional as F
+from torch.nn.modules import module as module_hooks
+
+
+class UnsupportedCandidateHead(ValueError):
+    """The head's semantics cannot be reproduced by plain selected-row linear."""
+
+
+def _check_head(head):
+    if type(head) is not nn.Linear:
+        raise UnsupportedCandidateHead("only exact torch.nn.Linear is supported")
+    if (
+        head.forward.__func__ is not nn.Linear.forward
+        if hasattr(head.forward, "__func__")
+        else True
+    ):
+        raise UnsupportedCandidateHead("overridden forward is unsupported")
+    if head._modules or head._buffers or set(head._parameters) != {"weight", "bias"}:
+        raise UnsupportedCandidateHead(
+            "head contains extra modules, buffers or parameters"
+        )
+    for name in (
+        "_forward_hooks",
+        "_forward_pre_hooks",
+        "_backward_hooks",
+        "_backward_pre_hooks",
+    ):
+        if getattr(head, name, None) or getattr(module_hooks, "_global" + name, None):
+            raise UnsupportedCandidateHead("module hooks would be bypassed")
+    for value in (head.weight, head.bias):
+        if value is None:
+            continue
+        if (
+            type(value) is not nn.Parameter
+            or value.is_quantized
+            or value.layout != torch.strided
+            or not value.is_floating_point()
+            or value.device.type == "meta"
+        ):
+            raise UnsupportedCandidateHead(
+                "requires ordinary dense floating-point Parameters"
+            )
+    if head.weight is None or head.weight.shape != (
+        head.out_features,
+        head.in_features,
+    ):
+        raise UnsupportedCandidateHead("invalid dense weight shape")
+    if head.bias is not None and (
+        head.bias.shape != (head.out_features,)
+        or head.bias.dtype != head.weight.dtype
+        or head.bias.device != head.weight.device
+    ):
+        raise UnsupportedCandidateHead(
+            "bias shape, dtype or device does not match weight"
+        )
+
+
+def candidate_logits(head, hidden_states, token_ids):
+    """Return [..., K] logits in supplied token order, including repeated IDs.
+
+    Dtype/autocast follow F.linear; no explicit precision conversion or detach.
+    Shape-dependent floating GEMM rounding may differ from full-vocabulary GEMM.
+    This validates indices, not tokenization, semantic labels, or probability mass.
+    """
+    _check_head(head)
+    if type(hidden_states) not in (torch.Tensor, nn.Parameter):
+        raise TypeError("hidden_states must be an ordinary Tensor")
+    if (
+        hidden_states.ndim < 1
+        or hidden_states.shape[-1] != head.in_features
+        or not hidden_states.is_floating_point()
+        or hidden_states.layout != torch.strided
+        or hidden_states.device != head.weight.device
+    ):
+        raise ValueError(
+            "hidden_states shape, floating layout or device does not match head"
+        )
+    if type(token_ids) is torch.Tensor:
+        if (
+            token_ids.ndim != 1
+            or token_ids.dtype != torch.long
+            or token_ids.device.type == "meta"
+        ):
+            raise ValueError("token_ids must be a one-dimensional int64 tensor")
+        indices = token_ids.to(device=head.weight.device)
+    elif isinstance(token_ids, (list, tuple)):
+        if any(type(value) is not int for value in token_ids):
+            raise TypeError("token IDs must be integers, not booleans or floats")
+        if any(value < 0 or value >= head.out_features for value in token_ids):
+            raise ValueError("token ID outside vocabulary")
+        indices = torch.tensor(token_ids, dtype=torch.long, device=head.weight.device)
+    else:
+        raise TypeError("token_ids must be a list, tuple or int64 tensor")
+    if not indices.numel():
+        raise ValueError("at least one candidate token is required")
+    if bool(((indices < 0) | (indices >= head.out_features)).any()):
+        raise ValueError("token ID outside vocabulary")
+    weight = head.weight.index_select(0, indices)
+    bias = None if head.bias is None else head.bias.index_select(0, indices)
+    return F.linear(hidden_states, weight, bias)

4B-5949/openjet_runtime/contracts.py

@@ -0,0 +1,102 @@
+"""Small shared contract. Prompts use a strict whitelist of input fields."""
+
+import json
+import math
+
+PROMPT_VERSION = "jev.dynamic.prompt.v2"
+LABELS = tuple("ABCDEFGHIJKLMNOP")
+BINARY_CRITERIA = [
+    {"id": "yes", "description": "The stated proposition is true."},
+    {"id": "no", "description": "The stated proposition is false."},
+]
+
+
+def validate_request(record):
+    for key in ("id", "group_id", "state", "instructions"):
+        if not isinstance(record.get(key), str) or not record[key].strip():
+            raise ValueError(f"{key} must be a nonempty string")
+    if record.get("primitive") not in ("choice", "noul", "score_level"):
+        raise ValueError("unsupported primitive")
+    criteria = record.get("criteria")
+    if not isinstance(criteria, list) or not 2 <= len(criteria) <= len(LABELS):
+        raise ValueError("criteria must contain 2..16 candidates")
+    ids = []
+    for candidate in criteria:
+        if not isinstance(candidate, dict):
+            raise TypeError("candidate must be an object")
+        for key in ("id", "description"):
+            if not isinstance(candidate.get(key), str) or not candidate[key].strip():
+                raise ValueError(f"candidate {key} must be nonempty")
+        ids.append(candidate["id"])
+    if len(set(ids)) != len(ids):
+        raise ValueError("duplicate candidate ids")
+    if record["primitive"] != "choice" and criteria != BINARY_CRITERIA:
+        raise ValueError("noul and score_level require canonical yes/no criteria")
+
+
+def validate_record(record):
+    validate_request(record)
+    if record.get("gold") not in [c["id"] for c in record["criteria"]]:
+        raise ValueError("gold must be a candidate id")
+    if not isinstance(record.get("provenance"), dict):
+        raise TypeError("provenance must be an object")
+
+
+def label_mapping(record):
+    validate_request(record)
+    return dict(zip(LABELS, (c["id"] for c in record["criteria"])))
+
+
+def render_prompt_parts(record):
+    """Text prefix/suffix; callers MUST check tokenizer boundary equivalence."""
+    validate_request(record)
+    prefix = "Shared state:\n" + record["state"] + "\n\n"
+    task = {
+        "primitive": record["primitive"],
+        "instructions": record["instructions"],
+        "criteria": [
+            {"label": label, "description": candidate["description"]}
+            for label, candidate in zip(LABELS, record["criteria"])
+        ],
+    }
+    suffix = json.dumps(task, ensure_ascii=False, sort_keys=True)
+    suffix += (
+        "\nReturn only the selected letter: "
+        + ", ".join(LABELS[: len(record["criteria"])])
+        + ".\nAnswer:"
+    )
+    return prefix, suffix
+
+
+def render_prompt(record):
+    return "".join(render_prompt_parts(record))
+
+
+def to_messages(record):
+    validate_record(record)
+    inverse = {candidate: label for label, candidate in label_mapping(record).items()}
+    return {
+        "messages": [
+            {"role": "user", "content": render_prompt(record)},
+            {"role": "assistant", "content": inverse[record["gold"]]},
+        ]
+    }
+
+
+def format_response(record, probabilities):
+    """Map ordered candidate probabilities; score_level is NOT aggregate Score."""
+    mapping = label_mapping(record)
+    values = list(probabilities)
+    if len(values) != len(mapping) or any(
+        not math.isfinite(p) or p < 0 or p > 1 for p in values
+    ):
+        raise ValueError("invalid probabilities")
+    if not math.isclose(sum(values), 1, abs_tol=1e-5):
+        raise ValueError("probabilities must sum to one")
+    distribution = dict(zip(mapping.values(), values))
+    result = {"type": record["primitive"], "probabilities": distribution}
+    if record["primitive"] == "choice":
+        result["choice"] = max(distribution, key=distribution.get)
+    else:
+        result["yes_probability"] = distribution["yes"]
+    return result

4B-5949/openjet_runtime/early_exit.py

@@ -0,0 +1,209 @@
+"""Actual Q1 no-cache layer-prefix execution for native Qwen3.5 decisions.
+
+Mirrors the mask/position preparation of Transformers Qwen3_5TextModel (5.16.1).
+This is a version-audited reference, not a generic model or cache implementation.
+"""
+
+from dataclasses import dataclass, replace
+
+import torch
+from torch import Tensor, nn
+from transformers.masking_utils import (
+    create_causal_mask,
+    create_recurrent_attention_mask,
+)
+
+from .candidate_projection import (
+    UnsupportedCandidateHead,
+    candidate_logits,
+)
+
+
+@dataclass(frozen=True)
+class DepthContinuation:
+    hidden: Tensor  # complete sequence residual, BEFORE final norm
+    position_ids: Tensor
+    position_embeddings: tuple[Tensor, Tensor]
+    masks: dict[str, Tensor | None]
+    depth: int
+    owner: object
+    model_signature: tuple
+
+
+@dataclass
+class DepthDecision:
+    depth: int
+    logits: Tensor  # [1,C]
+    projection_mode: str
+
+
+class QwenEarlyExit(nn.Module):
+    """Begin once, stop at a real depth, optionally continue without replay.
+
+    Q1 means one unpadded complete input sequence. No cache or token generation.
+    Continuations are ephemeral: do not mutate parameters/train-mode between
+    begin/advance/readout, or persist them across optimizer steps.
+    """
+
+    def __init__(self, model: nn.Module):
+        super().__init__()
+        self.model = model
+        if self.base.config.model_type not in {"qwen3_5", "qwen3_5_text"}:
+            raise ValueError("only Qwen3.5 text/conditional models are supported")
+        if len(self.backbone.layers) != self.backbone.config.num_hidden_layers:
+            raise ValueError("layer count/config mismatch")
+        if not set(self.backbone.config.layer_types) <= {
+            "linear_attention",
+            "full_attention",
+        }:
+            raise ValueError("unsupported hybrid layer type")
+        self._owner = object()
+
+    @property
+    def base(self):
+        return (
+            self.model.get_base_model()
+            if hasattr(self.model, "get_base_model")
+            else self.model
+        )
+
+    @property
+    def backbone(self):
+        return (
+            self.base.model.language_model
+            if self.base.config.model_type == "qwen3_5"
+            else self.base.model
+        )
+
+    @property
+    def full_depth(self):
+        return len(self.backbone.layers)
+
+    def _signature(self):
+        # Reference guard: optimizer updates and mode/device changes invalidate
+        # all outstanding continuations. No .data mutation is supported.
+        return (
+            tuple((id(module), module.training) for module in self.model.modules()),
+            tuple(
+                (id(parameter), parameter._version, parameter.device, parameter.dtype)
+                for parameter in self.model.parameters()
+            ),
+        )
+
+    def begin(self, input_ids: Tensor) -> DepthContinuation:
+        if (
+            input_ids.dtype != torch.long
+            or input_ids.ndim != 2
+            or input_ids.shape[0] != 1
+            or input_ids.shape[1] < 1
+        ):
+            raise ValueError("Q1 requires nonempty int64 input_ids [1,S], no padding")
+        for name in (
+            "image_token_id",
+            "video_token_id",
+            "vision_start_token_id",
+            "vision_end_token_id",
+        ):
+            token = getattr(self.base.config, name, None)
+            if token is not None and (input_ids == token).any():
+                raise ValueError("multimodal placeholders are unsupported")
+        embedding = self.base.get_input_embeddings()
+        if ((input_ids < 0) | (input_ids >= embedding.weight.shape[0])).any():
+            raise ValueError("input token outside vocabulary")
+        hidden = embedding(input_ids)
+        positions = (
+            torch.arange(input_ids.shape[1], device=hidden.device)
+            .view(1, 1, -1)
+            .expand(4, 1, -1)
+        )
+        text_positions = positions[0]
+        kwargs = {
+            "config": self.backbone.config,
+            "inputs_embeds": hidden,
+            "attention_mask": None,
+            "past_key_values": None,
+            "position_ids": text_positions,
+        }
+        masks = {
+            "full_attention": create_causal_mask(**kwargs),
+            "linear_attention": create_recurrent_attention_mask(**kwargs),
+        }
+        rotary = self.backbone.rotary_emb(hidden, positions[1:])
+        return DepthContinuation(
+            hidden, text_positions, rotary, masks, 0, self._owner, self._signature()
+        )
+
+    def _check_state(self, state):
+        if state.owner is not self._owner:
+            raise ValueError("continuation belongs to a different executor")
+        if state.model_signature != self._signature():
+            raise ValueError(
+                "stale continuation: model parameters or training mode changed"
+            )
+
+    def advance(self, state: DepthContinuation, target_depth: int) -> DepthContinuation:
+        self._check_state(state)
+        if (
+            type(target_depth) is not int
+            or not state.depth < target_depth <= self.full_depth
+        ):
+            raise ValueError("target depth must advance within the model")
+        hidden = state.hidden
+        for index in range(state.depth, target_depth):
+            hidden = self.backbone.layers[index](
+                hidden,
+                position_embeddings=state.position_embeddings,
+                attention_mask=state.masks[self.backbone.config.layer_types[index]],
+                position_ids=state.position_ids,
+                past_key_values=None,
+                use_cache=False,
+            )
+        return replace(state, hidden=hidden, depth=target_depth)
+
+    def readout(
+        self, state: DepthContinuation, candidate_token_ids: Tensor
+    ) -> DepthDecision:
+        self._check_state(state)
+        if state.depth < 1:
+            raise ValueError("readout requires at least one executed layer")
+        head = self.base.get_output_embeddings()
+        if (
+            candidate_token_ids.dtype != torch.long
+            or candidate_token_ids.ndim != 1
+            or candidate_token_ids.numel() < 2
+        ):
+            raise ValueError("require at least two int64 candidate tokens [C]")
+        if candidate_token_ids.device != state.hidden.device:
+            raise ValueError("candidate IDs must share hidden device")
+        if candidate_token_ids.unique().numel() != candidate_token_ids.numel():
+            raise ValueError("duplicate candidate token")
+        if (
+            (candidate_token_ids < 0) | (candidate_token_ids >= head.weight.shape[0])
+        ).any():
+            raise ValueError("candidate token outside vocabulary")
+        # Never replace the residual used by continuation with normalized hidden.
+        hidden = self.backbone.norm(state.hidden[:, -1])
+        try:
+            logits = candidate_logits(head, hidden, candidate_token_ids)
+            mode = "candidate_rows"
+        except UnsupportedCandidateHead:
+            # Preserve adapters/hooks/parametrizations by executing the real head.
+            logits = head(hidden).index_select(-1, candidate_token_ids)
+            mode = "full_head_fallback"
+        return DepthDecision(state.depth, logits, mode)
+
+    def forward(
+        self, input_ids: Tensor, candidate_token_ids: Tensor, *, depths: tuple[int, ...]
+    ):
+        if (
+            not depths
+            or any(type(d) is not int or not 1 <= d <= self.full_depth for d in depths)
+            or list(depths) != sorted(set(depths))
+        ):
+            raise ValueError("depths must be strictly increasing valid layer counts")
+        state = self.begin(input_ids)
+        decisions = []
+        for depth in depths:
+            state = self.advance(state, depth)
+            decisions.append(self.readout(state, candidate_token_ids))
+        return tuple(decisions)

4B-5949/openjet_runtime/runtime.py

@@ -0,0 +1,176 @@
+"""Portable, single-request merged Qwen3.5 decision reference runtime."""
+
+import json
+from pathlib import Path
+
+import torch
+import transformers
+
+from .contracts import PROMPT_VERSION, format_response, label_mapping, render_prompt
+from .early_exit import QwenEarlyExit
+
+
+class OpenJet:
+    """Explicit low/high, not automatic routing. Text-only; never truncates."""
+
+    def __init__(self, model, tokenizer, depth_config, max_length=8192):
+        if transformers.__version__ != "5.16.1":
+            raise RuntimeError("Layer execution is audited for transformers==5.16.1")
+        self.model = model.eval()
+        self.tokenizer = tokenizer
+        self.wrapper = QwenEarlyExit(self.model)
+        self.device = self.model.get_input_embeddings().weight.device
+        self.max_length = max_length
+        if type(max_length) is not int or not 1 <= max_length <= 8192:
+            raise ValueError("max_length must be an integer in 1..8192")
+        if depth_config.get("prompt_version") != PROMPT_VERSION:
+            raise ValueError("checkpoint prompt version does not match runtime")
+        full = depth_config.get("full_depth")
+        low = depth_config.get("exit_depth")
+        if full != self.wrapper.full_depth:
+            raise ValueError("checkpoint depth and model depth disagree")
+        if type(low) is not int or not 0 < low < full:
+            raise ValueError("checkpoint does not declare a trained shallow exit")
+        if self.wrapper.backbone.config.layer_types[low - 1] != "full_attention":
+            raise ValueError("shallow exit must be a full-attention boundary")
+        self.depths = {"low": low, "high": full}
+
+    @classmethod
+    def from_pretrained(cls, directory, device="cuda:0", dtype="bfloat16"):
+        """Load a local HF snapshot (download explicitly with a pinned revision)."""
+        from transformers import AutoConfig, AutoModelForCausalLM, AutoTokenizer
+
+        directory = Path(directory)
+        if (directory / "adapter_config.json").exists():
+            raise ValueError("Expected a merged snapshot, not an adapter directory")
+        if dtype not in ("float32", "bfloat16"):
+            raise ValueError("supported dtypes: float32, bfloat16")
+        config = AutoConfig.from_pretrained(directory, local_files_only=True)
+        loader = AutoModelForCausalLM
+        if config.model_type == "qwen3_5":
+            from transformers import Qwen3_5ForConditionalGeneration
+
+            loader = Qwen3_5ForConditionalGeneration
+        elif config.model_type != "qwen3_5_text":
+            raise ValueError("Expected Qwen3.5 text or conditional-generation model")
+        model = loader.from_pretrained(
+            directory,
+            local_files_only=True,
+            dtype=getattr(torch, dtype),
+            attn_implementation="sdpa",
+        ).to(device)
+        tokenizer = AutoTokenizer.from_pretrained(directory, local_files_only=True)
+        depth_config = json.loads((directory / "depth_config.json").read_text())
+        return cls(model, tokenizer, depth_config)
+
+    def _depth(self, effort):
+        if effort not in self.depths:
+            raise ValueError("effort must be low or high")
+        return self.depths[effort]
+
+    def compile(self, request):
+        """Exact chat/no-thinking contract used in original native evaluation."""
+        mapping = label_mapping(request)
+        prompt = self.tokenizer.apply_chat_template(
+            [{"role": "user", "content": render_prompt(request)}],
+            tokenize=False,
+            add_generation_prompt=True,
+            enable_thinking=False,
+        )
+        ids = self.tokenizer.encode(prompt, add_special_tokens=False)
+        if not ids or len(ids) > self.max_length:
+            raise ValueError("input exceeds runtime limit; no truncation permitted")
+        for name in (
+            "image_token_id",
+            "video_token_id",
+            "vision_start_token_id",
+            "vision_end_token_id",
+        ):
+            token = getattr(self.model.config, name, None)
+            if token is not None and token in ids:
+                raise ValueError("multimodal placeholders are unsupported")
+        tokens = []
+        for label in mapping:
+            token = self.tokenizer.encode(label, add_special_tokens=False)
+            joint = self.tokenizer.encode(prompt + label, add_special_tokens=False)
+            if len(token) != 1 or joint != ids + token:
+                raise ValueError(
+                    "candidate label is not single-token at answer boundary"
+                )
+            if token[0] in self.tokenizer.all_special_ids:
+                raise ValueError("candidate label must not be special token")
+            tokens.append(token[0])
+        if len(set(tokens)) != len(tokens):
+            raise ValueError("candidate token IDs must be unique")
+        return ids, tokens
+
+    @torch.inference_mode()
+    def decide(self, request, effort="high"):
+        depth = self._depth(effort)
+        ids, candidates = self.compile(request)
+        input_ids = torch.tensor([ids], dtype=torch.long, device=self.device)
+        candidate_ids = torch.tensor(candidates, dtype=torch.long, device=self.device)
+        if effort == "high":
+            # Match the original reference high/full-vocabulary head path.
+            output = self.model(
+                input_ids=input_ids,
+                attention_mask=torch.ones_like(input_ids),
+                position_ids=torch.arange(len(ids), device=self.device).unsqueeze(0),
+                past_key_values=None,
+                use_cache=True,
+                return_dict=True,
+                logits_to_keep=1,
+            )
+            logits = output.logits[0, -1].index_select(0, candidate_ids).float()
+            projection = "full_head"
+        else:
+            (decision,) = self.wrapper(input_ids, candidate_ids, depths=(depth,))
+            logits = decision.logits[0].float()
+            projection = decision.projection_mode
+        response = format_response(request, logits.softmax(-1).tolist())
+        response.update(
+            effort=effort,
+            executed_layers=depth,
+            prompt_tokens=len(ids),
+            logits=logits.tolist(),
+            projection=projection,
+            calibrated=False,
+        )
+        return response
+
+    @torch.inference_mode()
+    def generate_text(self, user_text, effort="high", max_new_tokens=128):
+        """TYPE greedy reference; replays prefix each token, not optimized serving."""
+        depth = self._depth(effort)
+        if not isinstance(user_text, str) or not user_text.strip():
+            raise ValueError("user_text must be nonempty")
+        if type(max_new_tokens) is not int or max_new_tokens < 1:
+            raise ValueError("max_new_tokens must be positive integer")
+        ids = self.tokenizer.apply_chat_template(
+            [{"role": "user", "content": user_text}],
+            tokenize=True,
+            add_generation_prompt=True,
+            enable_thinking=False,
+            return_dict=False,
+        )
+        if not ids or len(ids) + max_new_tokens > self.max_length:
+            raise ValueError("prompt plus generation reservation exceeds limit")
+        eos = self.model.generation_config.eos_token_id
+        eos = [eos] if isinstance(eos, int) else list(eos or [])
+        generated = []
+        for _ in range(max_new_tokens):
+            tensor = torch.tensor([ids + generated], device=self.device)
+            state = self.wrapper.advance(self.wrapper.begin(tensor), depth)
+            hidden = self.wrapper.backbone.norm(state.hidden[:, -1])
+            token = self.model.get_output_embeddings()(hidden)[0].argmax().item()
+            generated.append(token)
+            if token in eos:
+                break
+        return {
+            "text": self.tokenizer.decode(generated, skip_special_tokens=True),
+            "token_ids": generated,
+            "effort": effort,
+            "executed_layers_per_token": depth,
+            "finish_reason": "eos" if generated[-1] in eos else "length",
+            "prompt_tokens": len(ids),
+        }

4B-5949/processor_config.json

@@ -0,0 +1,61 @@
+{
+  "image_processor": {
+    "do_convert_rgb": true,
+    "do_normalize": true,
+    "do_rescale": true,
+    "do_resize": true,
+    "image_mean": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "image_processor_type": "Qwen2VLImageProcessor",
+    "image_std": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "merge_size": 2,
+    "patch_size": 16,
+    "resample": 3,
+    "rescale_factor": 0.00392156862745098,
+    "size": {
+      "longest_edge": 16777216,
+      "shortest_edge": 65536
+    },
+    "temporal_patch_size": 2
+  },
+  "processor_class": "Qwen3VLProcessor",
+  "video_processor": {
+    "do_convert_rgb": true,
+    "do_normalize": true,
+    "do_rescale": true,
+    "do_resize": true,
+    "do_sample_frames": true,
+    "fps": 2,
+    "image_mean": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "image_std": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "max_frames": 768,
+    "max_video_tokens": 768,
+    "merge_size": 2,
+    "min_frames": 4,
+    "patch_size": 16,
+    "resample": 3,
+    "rescale_factor": 0.00392156862745098,
+    "return_metadata": false,
+    "size": {
+      "longest_edge": 25165824,
+      "shortest_edge": 4096
+    },
+    "temporal_patch_size": 2,
+    "video_processor_type": "Qwen3VLVideoProcessor"
+  }
+}

4B-5949/provenance/BASE-LICENSE.txt

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2026 Alibaba Cloud
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

4B-5949/provenance/SOURCE-PROJECT-LICENSE.txt

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

4B-5949/provenance/code-license-provenance.json

@@ -0,0 +1,7 @@
+{
+  "source_project": "xDAN-ms-swift-jev",
+  "source_paths_and_hashes": "source-provenance.json",
+  "scope": "Source provenance only; assembly does not choose or grant a new license for exported code, adapters or training data.",
+  "source_project_license_sha256": "c71d239df91726fc519c6eb72d318ec65820627232b2f796219e87dcf35d0ab4",
+  "source_project_license_file": "SOURCE-PROJECT-LICENSE.txt"
+}

4B-5949/provenance/identity.json

@@ -0,0 +1,17 @@
+{
+  "base_id": "Qwen/Qwen3.5-4B",
+  "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a",
+  "checkpoint_step": 5949,
+  "adapter_sha256": "71154f60ec72c55d2c6c6147b9cdda9cc9a52d46297f3074dded7cdc9f5bc344",
+  "adapter_config_sha256": "c6caee3818ca1f6c8539e47fac9b9fa818d28b61a4bbb05f6f8444c0dd639e45",
+  "official80_sha256": "b3374e82f0e605762d40ab6455449c9cb2d315804a175d1cda985cba9beded35",
+  "software": {
+    "torch": "2.8.0+cu128",
+    "transformers": "5.16.1",
+    "peft": "0.20.0"
+  },
+  "merge_arithmetic": "float32 CPU safe_merge then bfloat16 storage",
+  "inference_dtype": "bfloat16",
+  "attention": "sdpa",
+  "gpu": "NVIDIA RTX PRO 6000 Blackwell Server Edition"
+}

4B-5949/release-manifest.json

@@ -0,0 +1,193 @@
+{
+  "format": "openjet.merged-release.v1",
+  "size": "4B",
+  "checkpoint_step": 5949,
+  "base_id": "Qwen/Qwen3.5-4B",
+  "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a",
+  "numerical_comparison_passed": false,
+  "decision_parity_passed": true,
+  "decision_agreement": 160,
+  "artifact_reviewed_variant": false,
+  "assembly_stage": "final",
+  "weight_arithmetic_passed": true,
+  "runtime_validation_status": "passed",
+  "private_staging": true,
+  "fresh_hub_download_gpu_check": "not established by assembly; consult publication receipt",
+  "files": [
+    {
+      "path": "LICENSE",
+      "bytes": 11544,
+      "sha256": "bbedc3fda3305820b977265f01b8619d87570a6739de3a5582c3464840f1e57a"
+    },
+    {
+      "path": "README.md",
+      "bytes": 1924,
+      "sha256": "4071763e7fbd1a16910c7bd8d7b6751e09b999f98bcdbe1d2f118f6b10f0c40f"
+    },
+    {
+      "path": "RUNTIME.md",
+      "bytes": 5556,
+      "sha256": "d9397e883c8b8709297f394c7c65eb0ee6b17c0994902b6f6d518d924fd2c854"
+    },
+    {
+      "path": "chat_template.jinja",
+      "bytes": 7756,
+      "sha256": "a4aee8afcf2e0711942cf848899be66016f8d14a889ff9ede07bca099c28f715"
+    },
+    {
+      "path": "config.json",
+      "bytes": 2829,
+      "sha256": "4b53bdb886bb68d228841409c2db2f5d7eb670c28ae4376849f58d3bd014968d"
+    },
+    {
+      "path": "decision-release.json",
+      "bytes": 1486,
+      "sha256": "51f8cd280dae539bc8cf1055b6a6ae64b8dfb13064ec3d155cf40ebe9079c690"
+    },
+    {
+      "path": "depth_config.json",
+      "bytes": 320,
+      "sha256": "50a4f97bbb3589222d81285ff94b33efd0f048d159166f6b93f2700e633be3e0"
+    },
+    {
+      "path": "evaluation/runtime-smoke.json",
+      "bytes": 4858,
+      "sha256": "60a80902675971ff6a054377852b936eec90f49f466eb7b8b077eee7b979f230"
+    },
+    {
+      "path": "evaluation/weight-arithmetic.json",
+      "bytes": 25495,
+      "sha256": "567c5c0f85e9b1db314c58c896542eaf0753d0723fa94e61f134660e813ea18d"
+    },
+    {
+      "path": "examples.py",
+      "bytes": 2332,
+      "sha256": "982ced288cae6ccf70ba70174f3c485398a8e712597b5b85423ba9b84318bf39"
+    },
+    {
+      "path": "generation_config.json",
+      "bytes": 116,
+      "sha256": "e4b598e9544d7567b3ae288efd8417e5b1646c957609b614c639163139336d12"
+    },
+    {
+      "path": "merge-provenance.json",
+      "bytes": 682,
+      "sha256": "3dcce1f04722cfd98a30c060b96eb1fc6b8b1d321904486acf1c1b2693c66d3a"
+    },
+    {
+      "path": "merged-evaluation.json",
+      "bytes": 3399,
+      "sha256": "b653bb0672e530a01ac58ab1b4597f7a9c81ddc157db2b0fdc8cf64ae73d85ed"
+    },
+    {
+      "path": "model-00001-of-00003.safetensors",
+      "bytes": 3991298872,
+      "sha256": "c04bef62040612a3376c144014c194687cdc19b18c3a807ee1136b4a713cbb0a"
+    },
+    {
+      "path": "model-00002-of-00003.safetensors",
+      "bytes": 3979833152,
+      "sha256": "7891d5b76b686893680e9cc074c2e17a788ff0cb03f64cc2ae4150bd804299a2"
+    },
+    {
+      "path": "model-00003-of-00003.safetensors",
+      "bytes": 1107487880,
+      "sha256": "a88eecd668f83cad773bd67c9c7e6e466c1746d489c55d906b48398a6679db65"
+    },
+    {
+      "path": "model.safetensors.index.json",
+      "bytes": 66236,
+      "sha256": "cd67b86e2cd9d329167224ac0026be69b39e078ad3056fc59339ba0a19fe4845"
+    },
+    {
+      "path": "openjet_runtime/__init__.py",
+      "bytes": 52,
+      "sha256": "df6eb864cf6d0c2f512fb00f17eeaa11fd790afca33a0f7b2f609aeb1f2b3944"
+    },
+    {
+      "path": "openjet_runtime/candidate_projection.py",
+      "bytes": 4438,
+      "sha256": "84dc4746b5fb06ac6a9024dde3ba8414d901acf2a62d010b0d66f26acfaf74a6"
+    },
+    {
+      "path": "openjet_runtime/contracts.py",
+      "bytes": 3847,
+      "sha256": "d8e8e5270ecd6dab917d886dda2d684c24696b811faa968bb3c399ceef5e356a"
+    },
+    {
+      "path": "openjet_runtime/early_exit.py",
+      "bytes": 7862,
+      "sha256": "89b7751a927a6d1348a454fb8ee986395423ad5686d855fb3ccab255c95d2566"
+    },
+    {
+      "path": "openjet_runtime/runtime.py",
+      "bytes": 8081,
+      "sha256": "6e7b0b131cb14ab0d25cc8fd6c7fc41738b799cfe6de1ccdbae2d09d7c64313c"
+    },
+    {
+      "path": "processor_config.json",
+      "bytes": 1220,
+      "sha256": "bfbc24af59a3e73a9cd0653b8d4ae758dfaec4e3e6c15dfb1c6c8ee8d5683c85"
+    },
+    {
+      "path": "provenance/BASE-LICENSE.txt",
+      "bytes": 11544,
+      "sha256": "bbedc3fda3305820b977265f01b8619d87570a6739de3a5582c3464840f1e57a"
+    },
+    {
+      "path": "provenance/SOURCE-PROJECT-LICENSE.txt",
+      "bytes": 11357,
+      "sha256": "c71d239df91726fc519c6eb72d318ec65820627232b2f796219e87dcf35d0ab4"
+    },
+    {
+      "path": "provenance/code-license-provenance.json",
+      "bytes": 397,
+      "sha256": "33a6e44402d44b5febe1451ec4978523858fd4e91f5f101e61ad9b39e9b21f2f"
+    },
+    {
+      "path": "provenance/identity.json",
+      "bytes": 682,
+      "sha256": "3dcce1f04722cfd98a30c060b96eb1fc6b8b1d321904486acf1c1b2693c66d3a"
+    },
+    {
+      "path": "requirements.txt",
+      "bytes": 139,
+      "sha256": "a5ceffcb009f4fe48882b20ef19d161ce53678abb5234175f45a8c82c3c868c6"
+    },
+    {
+      "path": "runtime-validation.json",
+      "bytes": 334,
+      "sha256": "96d38fd523772afe5cc96e645fa2bc14d20e58a1920da3822e6c42b6cfeaf3e4"
+    },
+    {
+      "path": "source-provenance.json",
+      "bytes": 773,
+      "sha256": "a623caf50f658bee5c1ad5a084e486a945e288fa93cdb798f89a58d0c5e050c4"
+    },
+    {
+      "path": "tokenizer.json",
+      "bytes": 19989325,
+      "sha256": "06b9509352d2af50381ab2247e083b80d32d5c0aba91c272ca9ff729b6a0e523"
+    },
+    {
+      "path": "tokenizer_config.json",
+      "bytes": 1165,
+      "sha256": "792fa3f0cb88b111e54ef3134c873531008c4df471d108da17903426e308aa7b"
+    },
+    {
+      "path": "training.md",
+      "bytes": 2692,
+      "sha256": "30544dc7ca3119ff10e88507651a3275b62f61b7e888f3332eb8fab632caae6b"
+    }
+  ],
+  "family_documentation_update": {
+    "source_manifest_sha256": "9dd56a1d3ee0526a6a9f2136393188edfe3a61eda79febc00093d65c6ed1b787",
+    "scope": "README only; source weights, runtime and evaluation evidence unchanged"
+  },
+  "organization_documentation_update": {
+    "source_repo": "gump2049/APUS-OpenJev-v1",
+    "source_revision": "e7e3cc0b9c82b91380ec6595120b7ce8abd32fd7",
+    "previous_manifest_sha256": "e7f96d2b30cc3296aa975efd7b3bf8721b97990937c523fe4d78d8c605199fce",
+    "scope": "README download links only; all model and runtime bytes unchanged"
+  }
+}

4B-5949/requirements.txt

@@ -0,0 +1,5 @@
+# Install torch with the CUDA build matching the deployment host first.
+torch==2.8.0
+transformers==5.16.1
+safetensors>=0.6
+huggingface_hub

4B-5949/runtime-validation.json

@@ -0,0 +1,10 @@
+{
+  "status": "passed",
+  "decisions": 160,
+  "max_probability_delta": 0.0,
+  "no_jev_import": true,
+  "long_input_rejected": true,
+  "invalid_effort_rejected": true,
+  "text_scope": "Execution smoke only; not TYPE accuracy or speed validation",
+  "source_sha256": "60a80902675971ff6a054377852b936eec90f49f466eb7b8b077eee7b979f230"
+}

4B-5949/source-provenance.json

@@ -0,0 +1,17 @@
+{
+  "contracts.py": {
+    "source": "jev/dynamic/contracts.py",
+    "source_sha256": "d8e8e5270ecd6dab917d886dda2d684c24696b811faa968bb3c399ceef5e356a",
+    "export_sha256": "d8e8e5270ecd6dab917d886dda2d684c24696b811faa968bb3c399ceef5e356a"
+  },
+  "candidate_projection.py": {
+    "source": "jev/dynamic/engine/candidate_projection.py",
+    "source_sha256": "84dc4746b5fb06ac6a9024dde3ba8414d901acf2a62d010b0d66f26acfaf74a6",
+    "export_sha256": "84dc4746b5fb06ac6a9024dde3ba8414d901acf2a62d010b0d66f26acfaf74a6"
+  },
+  "early_exit.py": {
+    "source": "jev/dynamic/native/early_exit.py",
+    "source_sha256": "e6ce9e983df0fdaaa9fb9bdf4a6e1f37e0610708007648b2f6b0b9d36e35d2e3",
+    "export_sha256": "89b7751a927a6d1348a454fb8ee986395423ad5686d855fb3ccab255c95d2566"
+  }
+}

4B-5949/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:06b9509352d2af50381ab2247e083b80d32d5c0aba91c272ca9ff729b6a0e523
+size 19989325

4B-5949/tokenizer_config.json

@@ -0,0 +1,33 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": true,
+  "local_files_only": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "processor_class": "Qwen3VLProcessor",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}

4B-5949/training.md

@@ -0,0 +1,24 @@
+# This release: 4B, step 5949
+
+This is the completed 5949-step SFT endpoint.
+
+# Training provenance
+
+Both families use the registered 5949-record SFT curriculum (3898 parent groups), with 5949 maximum steps and one epoch. Each checkpoint's recorded training step, epoch, and original trainer-state hash are preserved in the separate [LoRA archive manifest](https://huggingface.co/gump2049/xDAN-openJet-LoRA-Checkpoints/blob/1e5557f923746031f8187b7daf299b9bee41cb3c/manifest.json) at fixed revision `1e5557f923746031f8187b7daf299b9bee41cb3c` (repository access required). This merged package's `release-manifest.json` inventories inference artifacts and does not contain that per-checkpoint trainer-state record; intermediate checkpoints did not finish the whole schedule. Randomized loader order means the step number alone is not a verified count of unique examples seen at an intermediate checkpoint.
+
+| Source | Scheduled records | Parent groups |
+|---|---:|---:|
+| Mind2Web browser Choice | 1798 | 671 |
+| Mind2Web browser TYPE | 158 | 125 |
+| HelpSteer3 principle | 1300 | 1300 |
+| SGD | 513 | 10 |
+| GoEmotions independent-attribute Score | 500 | 297 |
+| BoolQ | 600 | 600 |
+| MNLI | 1000 | 1000 |
+| Local counterfactual | 80 | 20 |
+
+Parent groups can overlap across browser Choice/TYPE. The 4B run initialized from a 427-record pilot adapter (weight SHA256 `0047e5f1f0c98f17da93def94032041997592e1a609ddcab58de41f4d30e3a38`), while the inspected 9B config records no origin adapter. Do not add pilot records to the registered schedule as if all were independent.
+
+Decision objective: `0.5 CE(low) + 0.5 CE(high) + 0.1 KL(P_high.detach || P_low)`. TYPE examples use full-depth text cross-entropy. LoRA r=8, alpha=16, dropout=0; learning rate 1e-4; batch size 1; seed 20260920; training max length 6144. The recorded compiled schedule maximum is 5845 tokens. Training settings and data/schedule hashes are retained in the separate [LoRA checkpoint depth configuration](https://huggingface.co/gump2049/xDAN-openJet-LoRA-Checkpoints/blob/1e5557f923746031f8187b7daf299b9bee41cb3c/4b/checkpoint-5949/depth_config.json) at fixed archive revision `1e5557f923746031f8187b7daf299b9bee41cb3c`. The merged package's `depth_config.json` contains only portable inference and identity metadata; it is not the full training configuration.
+
+This is SFT with a within-model distillation term; it does not prove RLCD, online reinforcement learning or teacher-model OPD occurred. The declared public sources contain multiple licensing regimes (including CC-BY, CC-BY-SA and mixed-source material); separate provenance and redistribution review remains necessary. Raw datasets are not part of this upload.

9B-3000/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2026 Alibaba Cloud
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

9B-3000/README.md

@@ -0,0 +1,33 @@
+# APUS-OpenJev-v1 · 9B
+
+A decision model for browser agents and business workflows. This directory contains standalone BF16 weights and a runtime with selectable `effort="low"` and `effort="high"` compute budgets.
+
+[Model family](../README.md) · [Architecture](../ARCHITECTURE.md) · [Runtime guide](RUNTIME.md)
+
+## Quick start
+
+Use a CUDA-capable PyTorch environment.
+
+```bash
+python -m pip install huggingface_hub
+hf auth login
+hf download apus-ailab/APUS-OpenJev-v1 \
+  --include "9B-3000/*" --local-dir ./APUS-OpenJev-v1
+cd ./APUS-OpenJev-v1/9B-3000
+python -m pip install -r requirements.txt
+python examples.py . --device cuda:0 --effort high
+```
+
+The included runtime provides compute-budget selection. Use `high` for text generation.
+
+## Evaluation
+
+With the full compute budget, this merged model scores **68/80 (85.00%)** on the [Frozen80 development panel](https://huggingface.co/datasets/gump2049/xDAN-openJet-Eval-Frozen80-20260921): browser action selection, principle-based judgment, evidence-based questions, natural language inference, and attribute decisions.
+
+This reused development panel is an engineering reference, not an independent benchmark. BF16 merging changes some candidate probabilities; decision thresholds require revalidation. See [evaluation results](merged-evaluation.json) and [runtime checks](evaluation/runtime-smoke.json) for details.
+
+## Provenance
+
+We thank the Qwen team for the [Qwen3.5-9B](https://huggingface.co/Qwen/Qwen3.5-9B) base model. Training details and source records are in [training.md](training.md); artifact hashes are in [release-manifest.json](release-manifest.json). Consult [LICENSE](LICENSE) and the [base-model](provenance/BASE-LICENSE.txt) and [source-project](provenance/SOURCE-PROJECT-LICENSE.txt) notices.
+
+**Authors:** gumpcheng ([https://huggingface.co/xDAN2099](https://huggingface.co/xDAN2099)), zhangxu, [APUS AI-LAB](https://github.com/APUS-AI-Lab)

9B-3000/RUNTIME.md

@@ -0,0 +1,55 @@
+# xDAN-openJet merged reference runtime
+
+本目录是可随 HF merged 仓库发布的完整 Python 源码闭包，不需要安装 ms-swift、PEFT 或原项目。发布选择为 **4B checkpoint-5949**、**9B checkpoint-3000** 与 **9B checkpoint-5949**；各目录必须保留自己的 `depth_config.json`、完整 Qwen config、tokenizer、chat template、generation config 和 merged safetensors。不得将不同 checkpoint 的权重或 shallow/full 结果拼在一起。
+
+## 运行
+
+使用 CUDA 对应 PyTorch 2.8.0 构建，安装 `requirements.txt`。运行依赖 Transformer 私有模型层接口，所以严格要求 `transformers==5.16.1`；版本升级需重做层级及数值验证。这是原生 PyTorch 单卡单请求参考实现，不是 vLLM 服务。
+
+先把发布仓库的**固定 commit**完整下载到本机目录。仓库根目录有本目录中的 `openjet_runtime/` 与 `examples.py` 时：
+
+```bash
+python -m pip install -r requirements.txt
+python examples.py ./model-snapshot --device cuda:0 --effort both
+python examples.py ./model-snapshot --device cuda:0 --effort high --text
+```
+
+若代码与权重同在下载快照根目录，进入快照后把 `./model-snapshot` 改为 `.`。
+
+```python
+from openjet_runtime import OpenJet
+from examples import decision_examples
+
+model = OpenJet.from_pretrained("./model-snapshot")
+two_candidates, sixteen_candidates = decision_examples()
+print(model.decide(two_candidates, effort="low"))
+print(model.decide(sixteen_candidates, effort="high"))
+print(model.generate_text("Return only the text: red shoes", effort="high", max_new_tokens=32))
+```
+
+`examples.py` 中两候选工作流、16 候选浏览器和 TYPE 是接口演示，不是声称模型已通过的 benchmark。需要针对业务设计 prompt 和验证答案。
+
+## 接口和执行语义
+
+- `decide(request, effort)` 输入字段与原 `jev.dynamic.prompt.v2` 相同：`id/group_id/state/instructions/primitive/criteria`；每个候选有非空 `id` 和 `description`，2–16 个，ID 不重复。标签为 A–P，编译器验证每个标签在真实回答边界恰为一个 token。没有 gold 输入需求。
+- `primitive="choice"` 返回 `choice` 与按输入顺序映射的 `probabilities`；`noul/score_level` 使用 `contracts.py` 中固定 Yes/No 候选，返回 `yes_probability`。`score_level` 是单个命题的判断，不能当作完整序数 Score API。
+- `effort="low"` 执行 `depth_config.exit_depth`（这两项发布预期16），共享原 LM final norm 与候选行投影；`high` 执行 `full_depth`（预期32），保持原评测中的标准模型前向+完整 LM head 路径。读取 config，不凭参数规模推测层数。
+- 所有输入采用 tokenizer 自带 chat template、`enable_thinking=False`，超过8192 token直接报错。不会默默截断 state、instructions 或候选。
+- `probabilities` 是在当前候选集上的相对 softmax，**未经概率校准**；合并不自动带来可信置信度或校准保证。
+- `generate_text` 为 TYPE 文本保留的贪心参考路径；每个 token 重新计算前缀，方便与原评测逐 token 核对，但不适合宣传 tokens/s。达到上限明确返回 `finish_reason="length"`。
+- `both` 示例分别调用两个 effort；没有自动路由、不承诺共享两次调用的前缀缓存。本次便携发布不包含 KV 广播引擎、vLLM 插件、TypeSafe HTTP server 或多模态输入能力。
+
+## 合并验收（GPU，不能用静态测试代替）
+
+1. 固定同一 base revision、adapter SHA、tokenizer、chat template、dtype、attention backend 和 Transformers 版本，记录 merge 前后权重身份。保留 adapter 原文件。
+2. 同进程/新进程分别加载 base+adapter 和 merged；比较固定2候选、16候选、长输入、80题面板两 effort 的 token IDs、候选排序、logits、probabilities 和最终ID。保存逐题差异与最大绝对差，不能只比 aggregate accuracy。
+3. BF16 merge 会舍入：不能预先声明 bitwise一致或把漂移简单解释为无害；应报告数值误差和所有预测翻转。若超过事先制定的容忍值，停止发布数值等价结论，考虑 FP32 merge/存储再独立评测。
+4. fresh reload 验证 `depth_config` 与实际层数、模块边界一致。用层 forward hooks 检查 low只执行浅层、high执行全层；hooks会触发候选头保守fallback，不拿该测量做性能报告。
+5. TYPE短样本比较 token序列和EOS；分别测试空输入、非法候选/重复ID、超长输入明确失败。
+6. 在干净环境固定 HF revision 下载，运行本目录例子和同一小面板。记录显存、依赖、GPU型号以及权重checksum。成功加载只是第一关，不能当作质量或吞吐验收。
+
+原始数值等价门结果：`False`，保留原结果；决策完全一致门：`True`，一致 `160/160`。独立运行时 GPU 验收状态：`passed`。若数值门失败，此 BF16 包作为独立重评版本，禁止直接迁移概率/拒答/路由阈值；见 `merged-evaluation.json`。
+
+## 源码来历
+
+`contracts.py`、`candidate_projection.py` 与 `early_exit.py` 从已有本地实现原样提取（最后一项仅调整相对 import）；`source-provenance.json` 记录源路径与两端 SHA256。`runtime.py` 是最小加载及接口层；high 与 TYPE 分别对应原 `HFDecisionEngine._forward_batch` 和 `package_eval.prefix_next_token` 的执行语义。采用现有模型类：`qwen3_5` → `Qwen3_5ForConditionalGeneration`；`qwen3_5_text` → `AutoModelForCausalLM`。没有新增学习参数。

9B-3000/chat_template.jinja

@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}

9B-3000/config.json

@@ -0,0 +1,109 @@
+{
+  "architectures": [
+    "Qwen3_5ForConditionalGeneration"
+  ],
+  "dtype": "bfloat16",
+  "image_token_id": 248056,
+  "model_type": "qwen3_5",
+  "text_config": {
+    "attention_bias": false,
+    "attention_dropout": 0.0,
+    "attn_output_gate": true,
+    "bos_token_id": null,
+    "dtype": "bfloat16",
+    "eos_token_id": 248044,
+    "full_attention_interval": 4,
+    "head_dim": 256,
+    "hidden_act": "silu",
+    "hidden_size": 4096,
+    "initializer_range": 0.02,
+    "intermediate_size": 12288,
+    "layer_types": [
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention"
+    ],
+    "linear_conv_kernel_dim": 4,
+    "linear_key_head_dim": 128,
+    "linear_num_key_heads": 16,
+    "linear_num_value_heads": 32,
+    "linear_value_head_dim": 128,
+    "mamba_ssm_dtype": "float32",
+    "max_position_embeddings": 262144,
+    "mlp_only_layers": [],
+    "model_type": "qwen3_5_text",
+    "mtp_num_hidden_layers": 1,
+    "mtp_use_dedicated_embeddings": false,
+    "num_attention_heads": 16,
+    "num_hidden_layers": 32,
+    "num_key_value_heads": 4,
+    "pad_token_id": null,
+    "partial_rotary_factor": 0.25,
+    "rms_norm_eps": 1e-06,
+    "rope_parameters": {
+      "mrope_interleaved": true,
+      "mrope_section": [
+        11,
+        11,
+        10
+      ],
+      "partial_rotary_factor": 0.25,
+      "rope_theta": 10000000,
+      "rope_type": "default"
+    },
+    "tie_word_embeddings": false,
+    "use_cache": true,
+    "vocab_size": 248320
+  },
+  "tie_word_embeddings": false,
+  "transformers_version": "5.16.1",
+  "video_token_id": 248057,
+  "vision_config": {
+    "deepstack_visual_indexes": [],
+    "depth": 27,
+    "dtype": "bfloat16",
+    "hidden_act": "gelu_pytorch_tanh",
+    "hidden_size": 1152,
+    "in_channels": 3,
+    "initializer_range": 0.02,
+    "intermediate_size": 4304,
+    "model_type": "qwen3_5_vision",
+    "num_heads": 16,
+    "num_position_embeddings": 2304,
+    "out_hidden_size": 4096,
+    "patch_size": 16,
+    "spatial_merge_size": 2,
+    "temporal_patch_size": 2
+  },
+  "vision_end_token_id": 248054,
+  "vision_start_token_id": 248053
+}

9B-3000/decision-release.json

@@ -0,0 +1,35 @@
+{
+  "passed": true,
+  "scope": "Independent BF16 merged variant, frozen80 decision parity; NOT a probability-equivalent replacement",
+  "post_observation_scope_amendment": true,
+  "original_numerical_gate_passed": false,
+  "original_gate": "identical160_argmax_and_max_probability_delta_le_0.05",
+  "original_comparison_sha256": "d2ea8a7af10f09fdfde166384ffb18fb5a19ebf1a1281c58cb72806a45878c41",
+  "decision_agreement": 160,
+  "total_decisions": 160,
+  "independent_questions": 80,
+  "probability_calibration_transfer_validated": false,
+  "automatic_routing_validated": false,
+  "required_followup": "Recalibrate all probability, rejection and routing thresholds; independently evaluate new tasks",
+  "depths": {
+    "16": {
+      "before_correct": 63,
+      "after_correct": 63,
+      "changed_decisions": [],
+      "max_probability_abs_difference": 0.08810508251190186,
+      "mean_probability_abs_difference": 0.0019464968157012663,
+      "max_logit_abs_difference": 0.265625,
+      "mean_logit_abs_difference": 0.02963104248046875
+    },
+    "32": {
+      "before_correct": 68,
+      "after_correct": 68,
+      "changed_decisions": [],
+      "max_probability_abs_difference": 0.062176525592803955,
+      "mean_probability_abs_difference": 0.001304240933347387,
+      "max_logit_abs_difference": 0.25,
+      "mean_logit_abs_difference": 0.03828125
+    }
+  },
+  "publication_requires": "Further portable-runtime160, weight arithmetic, full file hashes and fresh HF reload gates"
+}

9B-3000/depth_config.json

@@ -0,0 +1,10 @@
+{
+  "prompt_version": "jev.dynamic.prompt.v2",
+  "exit_depth": 16,
+  "full_depth": 32,
+  "model_series": "xDAN-openJet",
+  "checkpoint_step": 3000,
+  "source_depth_config_sha256": "b6cab8011c151a7a5a5fce574addd8a5eb3d20e14cb516a1fb80d69563b4b4c9",
+  "training_mode": "two_exit",
+  "automatic_routing_validated": false
+}

9B-3000/evaluation/runtime-smoke.json

@@ -0,0 +1,187 @@
+{
+  "status": "passed",
+  "runtime_sha256": {
+    "openjet_runtime/__init__.py": "df6eb864cf6d0c2f512fb00f17eeaa11fd790afca33a0f7b2f609aeb1f2b3944",
+    "openjet_runtime/runtime.py": "6e7b0b131cb14ab0d25cc8fd6c7fc41738b799cfe6de1ccdbae2d09d7c64313c",
+    "openjet_runtime/early_exit.py": "89b7751a927a6d1348a454fb8ee986395423ad5686d855fb3ccab255c95d2566",
+    "openjet_runtime/contracts.py": "d8e8e5270ecd6dab917d886dda2d684c24696b811faa968bb3c399ceef5e356a",
+    "openjet_runtime/candidate_projection.py": "84dc4746b5fb06ac6a9024dde3ba8414d901acf2a62d010b0d66f26acfaf74a6"
+  },
+  "decisions": 160,
+  "max_probability_delta": 0.0,
+  "no_jev_import": true,
+  "long_input_rejected": true,
+  "invalid_effort_rejected": true,
+  "synthetic_examples": [
+    {
+      "id": "example-binary",
+      "type": "choice",
+      "probabilities": {
+        "close": 0.9984512329101562,
+        "refund": 0.0015487611526623368
+      },
+      "choice": "close",
+      "effort": "low",
+      "executed_layers": 16,
+      "prompt_tokens": 103,
+      "logits": [
+        7.8125,
+        1.34375
+      ],
+      "projection": "candidate_rows",
+      "calibrated": false
+    },
+    {
+      "id": "example-binary",
+      "type": "choice",
+      "probabilities": {
+        "close": 0.999480664730072,
+        "refund": 0.0005193048273213208
+      },
+      "choice": "close",
+      "effort": "high",
+      "executed_layers": 32,
+      "prompt_tokens": 103,
+      "logits": [
+        20.75,
+        13.1875
+      ],
+      "projection": "full_head",
+      "calibrated": false
+    },
+    {
+      "id": "example-browser",
+      "type": "choice",
+      "probabilities": {
+        "click-1": 0.0071670678444206715,
+        "click-2": 0.0058954693377017975,
+        "click-3": 0.07841967791318893,
+        "click-4": 0.02626730129122734,
+        "click-5": 0.012456356547772884,
+        "click-6": 0.05142887309193611,
+        "click-7": 0.032435785979032516,
+        "click-8": 0.02796139568090439,
+        "click-9": 0.031437840312719345,
+        "click-10": 0.031437840312719345,
+        "click-11": 0.08090898394584656,
+        "click-12": 0.11409995704889297,
+        "click-13": 0.10887493193149567,
+        "click-14": 0.17128431797027588,
+        "click-15": 0.05648357421159744,
+        "click-16": 0.16344064474105835
+      },
+      "choice": "click-14",
+      "effort": "low",
+      "executed_layers": 16,
+      "prompt_tokens": 383,
+      "logits": [
+        -0.158203125,
+        -0.353515625,
+        2.234375,
+        1.140625,
+        0.39453125,
+        1.8125,
+        1.3515625,
+        1.203125,
+        1.3203125,
+        1.3203125,
+        2.265625,
+        2.609375,
+        2.5625,
+        3.015625,
+        1.90625,
+        2.96875
+      ],
+      "projection": "candidate_rows",
+      "calibrated": false
+    },
+    {
+      "id": "example-browser",
+      "type": "choice",
+      "probabilities": {
+        "click-1": 0.0001901919167721644,
+        "click-2": 0.00011535723751876503,
+        "click-3": 0.00012279713700991124,
+        "click-4": 0.0001901919167721644,
+        "click-5": 9.563451021676883e-05,
+        "click-6": 8.984030864667147e-05,
+        "click-7": 5.4490898037329316e-05,
+        "click-8": 0.0001391473924741149,
+        "click-9": 0.0005503385909833014,
+        "click-10": 0.0001391473924741149,
+        "click-11": 0.0024664464872330427,
+        "click-12": 0.9950355291366577,
+        "click-13": 0.00033379721571691334,
+        "click-14": 0.00017866877897176892,
+        "click-15": 0.00010836809815373272,
+        "click-16": 0.0001901919167721644
+      },
+      "choice": "click-12",
+      "effort": "high",
+      "executed_layers": 32,
+      "prompt_tokens": 383,
+      "logits": [
+        12.4375,
+        11.9375,
+        12.0,
+        12.4375,
+        11.75,
+        11.6875,
+        11.1875,
+        12.125,
+        13.5,
+        12.125,
+        15.0,
+        21.0,
+        13.0,
+        12.375,
+        11.875,
+        12.4375
+      ],
+      "projection": "full_head",
+      "calibrated": false
+    }
+  ],
+  "text_smoke": {
+    "low": {
+      "text": "<think>nesssss\u5730\u4e2d\u56fd\u5bb6\u5730...\u2026sX\u5d07\u5cf0\u5cf0...\u2026",
+      "token_ids": [
+        248068,
+        2022,
+        82,
+        753,
+        95852,
+        106947,
+        95852,
+        1076,
+        1873,
+        82,
+        55,
+        98390,
+        97462,
+        97462,
+        1076,
+        1873
+      ],
+      "effort": "low",
+      "executed_layers_per_token": 16,
+      "finish_reason": "length",
+      "prompt_tokens": 28
+    },
+    "high": {
+      "text": "red shoes\n",
+      "token_ids": [
+        1114,
+        14850,
+        248046,
+        198,
+        248044
+      ],
+      "effort": "high",
+      "executed_layers_per_token": 32,
+      "finish_reason": "eos",
+      "prompt_tokens": 28
+    }
+  },
+  "text_scope": "Execution smoke only; not TYPE accuracy or speed validation"
+}

9B-3000/evaluation/weight-arithmetic.json

@@ -0,0 +1,908 @@
+{
+  "status": "passed",
+  "base_mtp_keys_not_loaded_by_transformers": [
+    "mtp.fc.weight",
+    "mtp.layers.0.input_layernorm.weight",
+    "mtp.layers.0.mlp.down_proj.weight",
+    "mtp.layers.0.mlp.gate_proj.weight",
+    "mtp.layers.0.mlp.up_proj.weight",
+    "mtp.layers.0.post_attention_layernorm.weight",
+    "mtp.layers.0.self_attn.k_norm.weight",
+    "mtp.layers.0.self_attn.k_proj.weight",
+    "mtp.layers.0.self_attn.o_proj.weight",
+    "mtp.layers.0.self_attn.q_norm.weight",
+    "mtp.layers.0.self_attn.q_proj.weight",
+    "mtp.layers.0.self_attn.v_proj.weight",
+    "mtp.norm.weight",
+    "mtp.pre_fc_norm_embedding.weight",
+    "mtp.pre_fc_norm_hidden.weight"
+  ],
+  "size": "9B",
+  "targeted_weight_matrices": 176,
+  "adapter_tensors": 352,
+  "all_targeted_weights_exact": true,
+  "scope": "FP32 LoRA arithmetic followed by BF16 rounding; not forward/probability equivalence",
+  "adapter_sha256": "14dd3cbaa26ced2ba5237dfff3aad93af4350dd9a43b869c7ce62cc9dd38d03b",
+  "checks": [
+    {
+      "weight": "model.language_model.layers.0.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.0.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.0.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.0.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.0.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.1.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.1.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.1.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.1.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.1.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.10.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.10.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.10.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.10.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.10.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.11.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.11.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.11.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.11.self_attn.k_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.11.self_attn.o_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.11.self_attn.q_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.11.self_attn.v_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.12.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.12.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.12.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.12.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.12.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.13.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.13.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.13.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.13.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.13.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.14.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.14.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.14.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.14.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.14.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.15.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.15.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.15.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.15.self_attn.k_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.15.self_attn.o_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.15.self_attn.q_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.15.self_attn.v_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.16.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.16.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.16.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.16.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.16.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.17.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.17.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.17.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.17.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.17.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.18.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.18.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.18.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.18.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.18.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.19.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.19.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.19.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.19.self_attn.k_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.19.self_attn.o_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.19.self_attn.q_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.19.self_attn.v_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.2.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.2.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.2.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.2.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.2.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.20.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.20.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.20.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.20.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.20.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.21.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.21.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.21.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.21.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.21.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.22.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.22.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.22.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.22.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.22.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.23.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.23.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.23.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.23.self_attn.k_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.23.self_attn.o_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.23.self_attn.q_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.23.self_attn.v_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.24.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.24.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.24.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.24.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.24.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.25.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.25.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.25.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.25.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.25.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.26.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.26.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.26.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.26.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.26.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.27.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.27.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.27.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.27.self_attn.k_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.27.self_attn.o_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.27.self_attn.q_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.27.self_attn.v_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.28.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.28.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.28.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.28.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.28.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.29.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.29.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.29.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.29.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.29.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.3.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.3.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.3.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.3.self_attn.k_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.3.self_attn.o_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.3.self_attn.q_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.3.self_attn.v_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.30.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.30.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.30.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.30.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.30.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.31.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.31.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.31.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.31.self_attn.k_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.31.self_attn.o_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.31.self_attn.q_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.31.self_attn.v_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.4.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.4.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.4.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.4.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.4.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.5.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.5.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.5.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.5.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.5.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.6.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.6.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.6.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.6.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.6.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.7.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.7.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.7.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.7.self_attn.k_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.7.self_attn.o_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.7.self_attn.q_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.7.self_attn.v_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.8.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.8.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.8.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.8.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.8.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.9.linear_attn.in_proj_qkv.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.9.linear_attn.out_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.9.mlp.down_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.9.mlp.gate_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    },
+    {
+      "weight": "model.language_model.layers.9.mlp.up_proj.weight",
+      "equal": true,
+      "max_abs_difference": 0.0
+    }
+  ]
+}

9B-3000/examples.py

@@ -0,0 +1,61 @@
+"""Run against a local merged HF snapshot; no project-local dependencies."""
+
+import argparse
+import json
+
+from openjet_runtime import OpenJet
+
+
+def decision_examples():
+    binary = {
+        "id": "example-binary",
+        "group_id": "example-binary",
+        "primitive": "choice",
+        "state": "Order 731 has been delivered. The customer's message says thank you.",
+        "instructions": "Select the appropriate next workflow action.",
+        "criteria": [
+            {"id": "close", "description": "Close the resolved support ticket."},
+            {"id": "refund", "description": "Refund an undelivered order."},
+        ],
+    }
+    browser = {
+        "id": "example-browser",
+        "group_id": "example-browser",
+        "primitive": "choice",
+        "state": "A settings page has 16 visible buttons labeled Page 1 through Page 16.",
+        "instructions": "Navigate to Page 12 by choosing its matching button.",
+        "criteria": [
+            {"id": f"click-{i}", "description": f"Click the Page {i} button."}
+            for i in range(1, 17)
+        ],
+    }
+    return binary, browser
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("model", help="Local merged snapshot directory")
+    parser.add_argument("--device", default="cuda:0")
+    parser.add_argument("--dtype", choices=["float32", "bfloat16"], default="bfloat16")
+    parser.add_argument("--effort", choices=["low", "high", "both"], default="both")
+    parser.add_argument(
+        "--text", action="store_true", help="Also run slow TYPE reference"
+    )
+    args = parser.parse_args()
+    runtime = OpenJet.from_pretrained(args.model, args.device, args.dtype)
+    efforts = ("low", "high") if args.effort == "both" else (args.effort,)
+    for effort in efforts:
+        for request in decision_examples():
+            result = runtime.decide(request, effort)
+            print(json.dumps({"example": request["id"], **result}, ensure_ascii=False))
+        if args.text:
+            result = runtime.generate_text(
+                "Return only the literal text to type into a search box for 'red shoes'.",
+                effort=effort,
+                max_new_tokens=32,
+            )
+            print(json.dumps({"example": "browser-type", **result}, ensure_ascii=False))
+
+
+if __name__ == "__main__":
+    main()

9B-3000/generation_config.json

@@ -0,0 +1,6 @@
+{
+  "_from_model_config": true,
+  "eos_token_id": 248044,
+  "transformers_version": "5.16.1",
+  "use_cache": true
+}

9B-3000/merge-provenance.json

@@ -0,0 +1,17 @@
+{
+  "base_id": "Qwen/Qwen3.5-9B",
+  "base_revision": "c202236235762e1c871ad0ccb60c8ee5ba337b9a",
+  "checkpoint_step": 3000,
+  "adapter_sha256": "14dd3cbaa26ced2ba5237dfff3aad93af4350dd9a43b869c7ce62cc9dd38d03b",
+  "adapter_config_sha256": "a3d03e9dfd4a3895d3a163ae679f931d957633deda178370ece7ebc86a1515ab",
+  "official80_sha256": "b3374e82f0e605762d40ab6455449c9cb2d315804a175d1cda985cba9beded35",
+  "software": {
+    "torch": "2.8.0+cu128",
+    "transformers": "5.16.1",
+    "peft": "0.20.0"
+  },
+  "merge_arithmetic": "float32 CPU safe_merge then bfloat16 storage",
+  "inference_dtype": "bfloat16",
+  "attention": "sdpa",
+  "gpu": "NVIDIA RTX PRO 6000 Blackwell Server Edition"
+}

9B-3000/merged-evaluation.json

@@ -0,0 +1,83 @@
+{
+  "scope": "80 fixed requests, both explicit efforts; merged-model comparison, not independent generalization evidence",
+  "checkpoint_step": 3000,
+  "base_id": "Qwen/Qwen3.5-9B",
+  "base_revision": "c202236235762e1c871ad0ccb60c8ee5ba337b9a",
+  "comparison": {
+    "passed": false,
+    "gate": "identical160_argmax_and_max_probability_delta_le_0.05",
+    "depths": {
+      "16": {
+        "before_correct": 63,
+        "after_correct": 63,
+        "changed_decisions": [],
+        "max_probability_abs_difference": 0.08810508251190186,
+        "mean_probability_abs_difference": 0.0019464968157012663,
+        "max_logit_abs_difference": 0.265625,
+        "mean_logit_abs_difference": 0.02963104248046875
+      },
+      "32": {
+        "before_correct": 68,
+        "after_correct": 68,
+        "changed_decisions": [],
+        "max_probability_abs_difference": 0.062176525592803955,
+        "mean_probability_abs_difference": 0.001304240933347387,
+        "max_logit_abs_difference": 0.25,
+        "mean_logit_abs_difference": 0.03828125
+      }
+    },
+    "before_sha256": "28412c70d8b456b35bb884a2fba4e85801b0135fc7957190927994a56d268a6f",
+    "after_sha256": "3f01a868d4fba91a8a1749df37a87f10f0874823eec5a978291c93231773c365",
+    "verified_at_unix": 1789984423.7235599
+  },
+  "runtime_validation": {
+    "status": "passed",
+    "decisions": 160,
+    "max_probability_delta": 0.0,
+    "no_jev_import": true,
+    "long_input_rejected": true,
+    "invalid_effort_rejected": true,
+    "text_scope": "Execution smoke only; not TYPE accuracy or speed validation",
+    "source_sha256": "70d686828090965063966dddac6d8454cf981c6614a406c1647617986e534c5d"
+  },
+  "decision_release": {
+    "passed": true,
+    "scope": "Independent BF16 merged variant, frozen80 decision parity; NOT a probability-equivalent replacement",
+    "post_observation_scope_amendment": true,
+    "original_numerical_gate_passed": false,
+    "original_gate": "identical160_argmax_and_max_probability_delta_le_0.05",
+    "original_comparison_sha256": "d2ea8a7af10f09fdfde166384ffb18fb5a19ebf1a1281c58cb72806a45878c41",
+    "decision_agreement": 160,
+    "total_decisions": 160,
+    "independent_questions": 80,
+    "probability_calibration_transfer_validated": false,
+    "automatic_routing_validated": false,
+    "required_followup": "Recalibrate all probability, rejection and routing thresholds; independently evaluate new tasks",
+    "depths": {
+      "16": {
+        "before_correct": 63,
+        "after_correct": 63,
+        "changed_decisions": [],
+        "max_probability_abs_difference": 0.08810508251190186,
+        "mean_probability_abs_difference": 0.0019464968157012663,
+        "max_logit_abs_difference": 0.265625,
+        "mean_logit_abs_difference": 0.02963104248046875
+      },
+      "32": {
+        "before_correct": 68,
+        "after_correct": 68,
+        "changed_decisions": [],
+        "max_probability_abs_difference": 0.062176525592803955,
+        "mean_probability_abs_difference": 0.001304240933347387,
+        "max_logit_abs_difference": 0.25,
+        "mean_logit_abs_difference": 0.03828125
+      }
+    },
+    "publication_requires": "Further portable-runtime160, weight arithmetic, full file hashes and fresh HF reload gates"
+  },
+  "reviewed_variant": {},
+  "decision_parity_passed": true,
+  "decision_agreement": 160,
+  "weight_arithmetic_status": "passed",
+  "raw_inputs_included": false
+}

9B-3000/model-00001-of-00006.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dd63614f1dc80dce2d83be3f3e69af1f8a0ebc8add9b8e0abf3910f563e44344
+size 2034237568

9B-3000/model-00002-of-00006.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3fe0ee3d29088f7f84ac7ba8c9d7562b91e6a31841c5b585c71779e618519e99
+size 3999615808