Add oQ6e build of KAT-Coder-V2.5-Dev-VL with restored vision tower and MTP head

.gitattributes

@@ -33,3 +33,4 @@ 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
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,134 @@
+---
+license: apache-2.0
+language:
+- en
+- zh
+pipeline_tag: image-text-to-text
+library_name: mlx
+tags:
+- code
+- agent
+- agentic-coding
+- moe
+- coding
+- vision
+- mtp
+- speculative-decoding
+- omlx
+- apple-silicon
+base_model:
+- Kwaipilot/KAT-Coder-V2.5-Dev
+- Qwen/Qwen3.6-35B-A3B
+---
+
+# KAT-Coder-V2.5-Dev-VL-oQ6e-mtp
+
+[`Kwaipilot/KAT-Coder-V2.5-Dev`](https://huggingface.co/Kwaipilot/KAT-Coder-V2.5-Dev) with its
+**vision tower and MTP head restored** from the base model it was fine-tuned from, quantized to
+6-bit for Apple Silicon. **28 GB.**
+
+This is the 6-bit member of a set built from one bf16 master:
+
+| level | size | HumanEval pass@1 | MTP acceptance | tok/cycle |
+|---|---|---|---|---|
+| [oQ8e](https://huggingface.co/jason-schulz/KAT-Coder-V2.5-Dev-VL-oQ8e-mtp) | 36 GB | 0.8598 | 74.1% | 2.27 |
+| [oQ6e](https://huggingface.co/jason-schulz/KAT-Coder-V2.5-Dev-VL-oQ6e-mtp) | 28 GB | 0.8659 | 75.5% | 2.46 |
+| [oQ4e](https://huggingface.co/jason-schulz/KAT-Coder-V2.5-Dev-VL-oQ4e-mtp) | 20 GB | 0.8963 | 79.3% | 2.54 |
+
+**Read the accuracy column as a tie, not a ranking.** All three were measured the same day, same
+settings (full 164 problems, `batch_size=1`, thinking enabled, deterministic profile). The oQ8e
+build scored 146/164 on an earlier run and 141/164 on the run in this table — a 5-problem swing on
+an unchanged model, which is as large as the entire spread between levels. Speculative decoding plus
+Metal batching numerics means "deterministic" is not bit-reproducible here. The honest conclusion is
+that **8, 6, and 4-bit are indistinguishable on this benchmark**, not that 4-bit is better.
+
+> **Read this first: what you need to load it.**
+>
+> The `oQ6e` quantization is **not** a proprietary format. oQ writes standard mlx-lm safetensors
+> with a standard per-module quantization config, and portable oQ models load in stock MLX apps.
+>
+> **What makes *this repo* oMLX-only is the restored MTP head.** Stock `mlx-vlm` does a strict load
+> and rejects the 42 extra `language_model.mtp.*` tensors:
+> `ValueError: Received 42 parameters not in model`. So as published this needs
+> [oMLX](https://github.com/jundot/omlx) on Apple Silicon — for the MTP tensors, not the quant.
+>
+> Not loadable by llama.cpp, vLLM, or Transformers (MLX safetensors either way).
+
+## What was done
+
+| | source |
+|---|---|
+| text weights | `Kwaipilot/KAT-Coder-V2.5-Dev`, unmodified (bit-exact; only the MoE expert tensor **layout** was converted) |
+| vision tower (333 tensors) | `Qwen/Qwen3.6-35B-A3B` |
+| MTP head (19 tensors) | `Qwen/Qwen3.6-35B-A3B` |
+| quantization | oMLX `oQ6` enhanced, `group_size=64`, affine, bf16 scales, 128x512 imatrix |
+
+Kwaipilot's open-weight release ships language-model weights only — the vision components and the
+multi-token-prediction head were stripped. Both are recoverable because KAT's `config.json` is
+byte-identical to `Qwen/Qwen3.6-35B-A3B`'s except for one field, and the text architecture is
+unchanged.
+
+The tensor key set is **identical** to an oMLX `oQ8e-mtp` build of the stock base model — 2052
+tensors, with the MTP head correctly `switch_mlp`-named.
+
+## Verification
+
+Measured on this exact checkpoint:
+
+- **Loads as a VLM.** Server log reports `VLMBatchedEngine loaded`, not the silent VLM to LLM
+  fallback that yields a vision-less model.
+- **Vision.** 8-part probe (dense code screenshots, high-entropy strings, UI defect-spotting, mixed
+  heading sizes, sparse-glyph sweeps). Transcription is exact except for two failure modes the
+  **native Qwen3.6 base reproduces as well**: the capital `Q` in `sk-7Qv3...` reads as lowercase,
+  and a sparse 384x320 @64px image misreads. No regression against the oQ8e build.
+- **MTP.** 75.5% draft acceptance, 2.46 tokens/cycle over 8 generations.
+- **HumanEval.** 0.8659 pass@1 (142/164), 739.7s, thinking enabled.
+
+**Caveat on MTP at low bit depths.** oMLX warns that quantizing a preserved MTP head can collapse
+draft acceptance toward ~0%. That did not happen here, and the likely reason is that the imatrix was
+**recomputed with the MTP head present** rather than reused from a pre-graft cache. If you build
+your own level from a stale imatrix, check the `MTP[n] ... accept=` line before trusting the
+speedup — do not assume 4-bit is universally safe for MTP-preserved heads.
+
+## Reproducing
+
+Build scripts live in the oQ8e repo: [`build/merge_kat_vl.py`](https://huggingface.co/jason-schulz/KAT-Coder-V2.5-Dev-VL-oQ8e-mtp/blob/main/build/merge_kat_vl.py)
+and [`build/graft_mtp.py`](https://huggingface.co/jason-schulz/KAT-Coder-V2.5-Dev-VL-oQ8e-mtp/blob/main/build/graft_mtp.py), with full instructions in its
+[README](https://huggingface.co/jason-schulz/KAT-Coder-V2.5-Dev-VL-oQ8e-mtp#reproducing). Budget ~146 GB of free disk; peak RAM is about 8 GB.
+
+Steps 1 and 2 produce the bf16 master. This repo is step 3 with `"oq_level": 6`:
+
+```bash
+curl -X POST http://127.0.0.1:8899/admin/api/oq/start \
+  -H 'Content-Type: application/json' -d '{
+    "model_path": "'"$HOME"'/Models/_build/KAT-Coder-V2.5-Dev-VL",
+    "oq_level": 6, "enhanced": true, "dtype": "bfloat16", "group_size": 64,
+    "preserve_mtp": true }'
+```
+
+Then set `mtp_enabled` on the new model id — it is per-model and off by default, and a model with
+MTP disabled logs no acceptance line at all, which is easy to misread as a collapsed head.
+
+## Limitations
+
+- **oMLX only *as published*** — because of the MTP head, not the quantization. Strip the `mtp.*`
+  tensors and set `text_config.mtp_num_hidden_layers = 0` and the remainder is a portable oQ
+  checkpoint that loads in stock MLX apps; you lose speculative decoding, keep vision and the text
+  model.
+- **Vision is untrained for this checkpoint.** Good for screenshots of code/terminals/UI; unreliable
+  for isolated large glyphs and occasionally for single characters in random strings.
+- **Set a `max_tokens` cap.** Long reasoning excursions stall rather than error in an agent loop.
+- HumanEval is a narrow benchmark. "Indistinguishable on HumanEval" is not "indistinguishable in
+  agentic coding use" — if you have a workload that discriminates between these levels, that result
+  is more informative than the table above.
+- Chinese/English only, inherited from the base.
+
+## Credits
+
+- [Kwaipilot](https://huggingface.co/Kwaipilot) — KAT-Coder-V2.5-Dev
+- [Qwen](https://huggingface.co/Qwen) — Qwen3.6-35B-A3B, source of the vision tower and MTP head
+- [oMLX](https://github.com/jundot/omlx) — quantization and serving
+- The [KAT-Coder community discussions](https://huggingface.co/Kwaipilot/KAT-Coder-V2.5-Dev/discussions),
+  where the vision and MTP transplants were first proposed and tested
+
+Apache-2.0, inherited from both parents.

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 %}
+            {{- '<|im_start|>system\n' + content + '<|im_end|>' + '\n' }}
+        {%- 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 (preserve_thinking is defined and preserve_thinking is true) or (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 | string if args_value is string else args_value | tojson | safe %}
+                        {{- 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 %}

generation_config.json

@@ -0,0 +1,12 @@
+{
+    "bos_token_id": 248044,
+    "do_sample": true,
+    "eos_token_id": [
+        248046,
+        248044
+    ],
+    "pad_token_id": 248044,
+    "temperature": 1.0,
+    "top_k": 20,
+    "top_p": 0.95
+}

model-00001-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:16cca919033c94cf832a475a7fe5d2efafc460d40653572fecaaf0ce4a855983
+size 5004918171

model-00002-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:36292571def4a5d1320e92c59a545830fc1620fc0836a7ec2b472d74f5f18ff9
+size 5016397261

model-00003-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:925736ff1ef3658012e6dfcc2b5c355a0db2f4590038aaecffb4a27e61fc2e79
+size 5016397336

model-00004-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53e76823c19502dbf2e56fdd014d295288bb05489fd61a2e271a2fe0c6979ee6
+size 5016397338

model-00005-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:18cd818ed7bc5a577e18defe27b3a366c7845a183d1b7f85fb16cf3a5ffbe5b1
+size 5016397338

model-00006-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:398fb00481aac500f4aa94098f3c3cd6079fc37272c2af90046db62e8b9174b3
+size 5016397298

model-00007-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f0a4a8e032d155d26b828f1a677c6af910b420dbf95d4c9e6109d127183d31c6
+size 218104246

oq_imatrix_report.json

@@ -0,0 +1,588 @@
+{
+  "enabled": true,
+  "cache_path": "/Users/jasonschulz/Models/.oqe_imatrix/KAT-Coder-V2.5-Dev-VL-fb9cf1e47cc2-s128-l512.npz",
+  "cache_reused": false,
+  "entry_count": 523,
+  "calib_dataset": "oqe_code_multilingual",
+  "collection": {
+    "dataset": "oqe_code_multilingual",
+    "requested_samples": 128,
+    "seq_length": 512,
+    "adaptive": true,
+    "adaptive_step_samples": 128,
+    "adaptive_max_samples": 1024,
+    "available_samples": 1024,
+    "micro_batch_size": 16,
+    "micro_batches": 8,
+    "batch_plan": {
+      "micro_batch_size": 16,
+      "estimated_sample_bytes": 33554432,
+      "capture_budget_bytes": 805306368,
+      "system_available_bytes": 102044925952,
+      "metal_available_bytes": 130563094994,
+      "live_available_bytes": 102044925952,
+      "model_bytes": 20897700290,
+      "remaining_available_bytes": 81147225662,
+      "fits_one_sample": true,
+      "hidden_size": 2048,
+      "num_experts": 256,
+      "top_k": 8,
+      "gemma4_state_bytes": 0,
+      "num_hidden_layers": 40,
+      "num_kv_shared_layers": 0,
+      "per_layer_input_size": 0
+    },
+    "processed_samples": 128,
+    "installed_modules": 634,
+    "capture_module_classes": {
+      "Linear": 151,
+      "QuantizedLinear": 360,
+      "QuantizedSwitchLinear": 123
+    },
+    "switch_capture_modules": 123,
+    "requires_expert_counts": true,
+    "coverage_sufficient": true,
+    "collection_sufficient": true,
+    "coverage": {
+      "has_expert_counts": true,
+      "expert_modules": 123,
+      "total_experts": 31488,
+      "active_experts": 31488,
+      "zero_count_experts": 0,
+      "active_ratio": 1.0,
+      "min_count": 40,
+      "p05_count": 588.0,
+      "p10_count": 759.0,
+      "median_count": 1742.0,
+      "max_count": 36257,
+      "min_required_count": 16,
+      "required_percentile": 5
+    },
+    "rounds": [
+      {
+        "processed_samples": 128,
+        "coverage_sufficient": true,
+        "collection_sufficient": true,
+        "coverage": {
+          "has_expert_counts": true,
+          "expert_modules": 123,
+          "total_experts": 31488,
+          "active_experts": 31488,
+          "zero_count_experts": 0,
+          "active_ratio": 1.0,
+          "min_count": 40,
+          "p05_count": 588.0,
+          "p10_count": 759.0,
+          "median_count": 1742.0,
+          "max_count": 36257,
+          "min_required_count": 16,
+          "required_percentile": 5
+        }
+      }
+    ]
+  },
+  "expert_coverage": {
+    "has_expert_counts": true,
+    "expert_modules": 123,
+    "total_experts": 31488,
+    "active_experts": 31488,
+    "zero_count_experts": 0,
+    "active_ratio": 1.0,
+    "min_count": 40,
+    "p05_count": 588.0,
+    "p10_count": 759.0,
+    "median_count": 1742.0,
+    "max_count": 36257,
+    "min_required_count": 16,
+    "required_percentile": 5
+  },
+  "applied": [
+    "language_model.model.layers.0.linear_attn.in_proj_a",
+    "language_model.model.layers.0.linear_attn.in_proj_b",
+    "language_model.model.layers.0.linear_attn.in_proj_qkv",
+    "language_model.model.layers.0.linear_attn.in_proj_z",
+    "language_model.model.layers.0.linear_attn.out_proj",
+    "language_model.model.layers.0.mlp.shared_expert.down_proj",
+    "language_model.model.layers.0.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.0.mlp.shared_expert.up_proj",
+    "language_model.model.layers.0.mlp.shared_expert_gate",
+    "language_model.model.layers.0.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.0.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.0.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.1.linear_attn.in_proj_a",
+    "language_model.model.layers.1.linear_attn.in_proj_b",
+    "language_model.model.layers.1.linear_attn.in_proj_qkv",
+    "language_model.model.layers.1.linear_attn.in_proj_z",
+    "language_model.model.layers.1.linear_attn.out_proj",
+    "language_model.model.layers.1.mlp.shared_expert.down_proj",
+    "language_model.model.layers.1.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.1.mlp.shared_expert.up_proj",
+    "language_model.model.layers.1.mlp.shared_expert_gate",
+    "language_model.model.layers.1.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.1.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.1.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.10.linear_attn.in_proj_a",
+    "language_model.model.layers.10.linear_attn.in_proj_b",
+    "language_model.model.layers.10.linear_attn.in_proj_qkv",
+    "language_model.model.layers.10.linear_attn.in_proj_z",
+    "language_model.model.layers.10.linear_attn.out_proj",
+    "language_model.model.layers.10.mlp.shared_expert.down_proj",
+    "language_model.model.layers.10.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.10.mlp.shared_expert.up_proj",
+    "language_model.model.layers.10.mlp.shared_expert_gate",
+    "language_model.model.layers.10.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.10.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.10.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.11.mlp.shared_expert.down_proj",
+    "language_model.model.layers.11.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.11.mlp.shared_expert.up_proj",
+    "language_model.model.layers.11.mlp.shared_expert_gate",
+    "language_model.model.layers.11.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.11.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.11.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.11.self_attn.k_proj",
+    "language_model.model.layers.11.self_attn.o_proj",
+    "language_model.model.layers.11.self_attn.q_proj",
+    "language_model.model.layers.11.self_attn.v_proj",
+    "language_model.model.layers.12.linear_attn.in_proj_a",
+    "language_model.model.layers.12.linear_attn.in_proj_b",
+    "language_model.model.layers.12.linear_attn.in_proj_qkv",
+    "language_model.model.layers.12.linear_attn.in_proj_z",
+    "language_model.model.layers.12.linear_attn.out_proj",
+    "language_model.model.layers.12.mlp.shared_expert.down_proj",
+    "language_model.model.layers.12.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.12.mlp.shared_expert.up_proj",
+    "language_model.model.layers.12.mlp.shared_expert_gate",
+    "language_model.model.layers.12.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.12.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.12.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.13.linear_attn.in_proj_a",
+    "language_model.model.layers.13.linear_attn.in_proj_b",
+    "language_model.model.layers.13.linear_attn.in_proj_qkv",
+    "language_model.model.layers.13.linear_attn.in_proj_z",
+    "language_model.model.layers.13.linear_attn.out_proj",
+    "language_model.model.layers.13.mlp.shared_expert.down_proj",
+    "language_model.model.layers.13.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.13.mlp.shared_expert.up_proj",
+    "language_model.model.layers.13.mlp.shared_expert_gate",
+    "language_model.model.layers.13.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.13.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.13.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.14.linear_attn.in_proj_a",
+    "language_model.model.layers.14.linear_attn.in_proj_b",
+    "language_model.model.layers.14.linear_attn.in_proj_qkv",
+    "language_model.model.layers.14.linear_attn.in_proj_z",
+    "language_model.model.layers.14.linear_attn.out_proj",
+    "language_model.model.layers.14.mlp.shared_expert.down_proj",
+    "language_model.model.layers.14.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.14.mlp.shared_expert.up_proj",
+    "language_model.model.layers.14.mlp.shared_expert_gate",
+    "language_model.model.layers.14.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.14.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.14.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.15.mlp.shared_expert.down_proj",
+    "language_model.model.layers.15.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.15.mlp.shared_expert.up_proj",
+    "language_model.model.layers.15.mlp.shared_expert_gate",
+    "language_model.model.layers.15.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.15.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.15.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.15.self_attn.k_proj",
+    "language_model.model.layers.15.self_attn.o_proj",
+    "language_model.model.layers.15.self_attn.q_proj",
+    "language_model.model.layers.15.self_attn.v_proj",
+    "language_model.model.layers.16.linear_attn.in_proj_a",
+    "language_model.model.layers.16.linear_attn.in_proj_b",
+    "language_model.model.layers.16.linear_attn.in_proj_qkv",
+    "language_model.model.layers.16.linear_attn.in_proj_z",
+    "language_model.model.layers.16.linear_attn.out_proj",
+    "language_model.model.layers.16.mlp.shared_expert.down_proj",
+    "language_model.model.layers.16.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.16.mlp.shared_expert.up_proj",
+    "language_model.model.layers.16.mlp.shared_expert_gate",
+    "language_model.model.layers.16.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.16.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.16.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.17.linear_attn.in_proj_a",
+    "language_model.model.layers.17.linear_attn.in_proj_b",
+    "language_model.model.layers.17.linear_attn.in_proj_qkv",
+    "language_model.model.layers.17.linear_attn.in_proj_z",
+    "language_model.model.layers.17.linear_attn.out_proj",
+    "language_model.model.layers.17.mlp.shared_expert.down_proj",
+    "language_model.model.layers.17.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.17.mlp.shared_expert.up_proj",
+    "language_model.model.layers.17.mlp.shared_expert_gate",
+    "language_model.model.layers.17.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.17.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.17.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.18.linear_attn.in_proj_a",
+    "language_model.model.layers.18.linear_attn.in_proj_b",
+    "language_model.model.layers.18.linear_attn.in_proj_qkv",
+    "language_model.model.layers.18.linear_attn.in_proj_z",
+    "language_model.model.layers.18.linear_attn.out_proj",
+    "language_model.model.layers.18.mlp.shared_expert.down_proj",
+    "language_model.model.layers.18.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.18.mlp.shared_expert.up_proj",
+    "language_model.model.layers.18.mlp.shared_expert_gate",
+    "language_model.model.layers.18.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.18.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.18.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.19.mlp.shared_expert.down_proj",
+    "language_model.model.layers.19.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.19.mlp.shared_expert.up_proj",
+    "language_model.model.layers.19.mlp.shared_expert_gate",
+    "language_model.model.layers.19.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.19.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.19.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.19.self_attn.k_proj",
+    "language_model.model.layers.19.self_attn.o_proj",
+    "language_model.model.layers.19.self_attn.q_proj",
+    "language_model.model.layers.19.self_attn.v_proj",
+    "language_model.model.layers.2.linear_attn.in_proj_a",
+    "language_model.model.layers.2.linear_attn.in_proj_b",
+    "language_model.model.layers.2.linear_attn.in_proj_qkv",
+    "language_model.model.layers.2.linear_attn.in_proj_z",
+    "language_model.model.layers.2.linear_attn.out_proj",
+    "language_model.model.layers.2.mlp.shared_expert.down_proj",
+    "language_model.model.layers.2.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.2.mlp.shared_expert.up_proj",
+    "language_model.model.layers.2.mlp.shared_expert_gate",
+    "language_model.model.layers.2.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.2.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.2.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.20.linear_attn.in_proj_a",
+    "language_model.model.layers.20.linear_attn.in_proj_b",
+    "language_model.model.layers.20.linear_attn.in_proj_qkv",
+    "language_model.model.layers.20.linear_attn.in_proj_z",
+    "language_model.model.layers.20.linear_attn.out_proj",
+    "language_model.model.layers.20.mlp.shared_expert.down_proj",
+    "language_model.model.layers.20.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.20.mlp.shared_expert.up_proj",
+    "language_model.model.layers.20.mlp.shared_expert_gate",
+    "language_model.model.layers.20.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.20.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.20.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.21.linear_attn.in_proj_a",
+    "language_model.model.layers.21.linear_attn.in_proj_b",
+    "language_model.model.layers.21.linear_attn.in_proj_qkv",
+    "language_model.model.layers.21.linear_attn.in_proj_z",
+    "language_model.model.layers.21.linear_attn.out_proj",
+    "language_model.model.layers.21.mlp.shared_expert.down_proj",
+    "language_model.model.layers.21.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.21.mlp.shared_expert.up_proj",
+    "language_model.model.layers.21.mlp.shared_expert_gate",
+    "language_model.model.layers.21.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.21.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.21.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.22.linear_attn.in_proj_a",
+    "language_model.model.layers.22.linear_attn.in_proj_b",
+    "language_model.model.layers.22.linear_attn.in_proj_qkv",
+    "language_model.model.layers.22.linear_attn.in_proj_z",
+    "language_model.model.layers.22.linear_attn.out_proj",
+    "language_model.model.layers.22.mlp.shared_expert.down_proj",
+    "language_model.model.layers.22.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.22.mlp.shared_expert.up_proj",
+    "language_model.model.layers.22.mlp.shared_expert_gate",
+    "language_model.model.layers.22.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.22.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.22.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.23.mlp.shared_expert.down_proj",
+    "language_model.model.layers.23.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.23.mlp.shared_expert.up_proj",
+    "language_model.model.layers.23.mlp.shared_expert_gate",
+    "language_model.model.layers.23.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.23.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.23.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.23.self_attn.k_proj",
+    "language_model.model.layers.23.self_attn.o_proj",
+    "language_model.model.layers.23.self_attn.q_proj",
+    "language_model.model.layers.23.self_attn.v_proj",
+    "language_model.model.layers.24.linear_attn.in_proj_a",
+    "language_model.model.layers.24.linear_attn.in_proj_b",
+    "language_model.model.layers.24.linear_attn.in_proj_qkv",
+    "language_model.model.layers.24.linear_attn.in_proj_z",
+    "language_model.model.layers.24.linear_attn.out_proj",
+    "language_model.model.layers.24.mlp.shared_expert.down_proj",
+    "language_model.model.layers.24.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.24.mlp.shared_expert.up_proj",
+    "language_model.model.layers.24.mlp.shared_expert_gate",
+    "language_model.model.layers.24.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.24.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.24.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.25.linear_attn.in_proj_a",
+    "language_model.model.layers.25.linear_attn.in_proj_b",
+    "language_model.model.layers.25.linear_attn.in_proj_qkv",
+    "language_model.model.layers.25.linear_attn.in_proj_z",
+    "language_model.model.layers.25.linear_attn.out_proj",
+    "language_model.model.layers.25.mlp.shared_expert.down_proj",
+    "language_model.model.layers.25.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.25.mlp.shared_expert.up_proj",
+    "language_model.model.layers.25.mlp.shared_expert_gate",
+    "language_model.model.layers.25.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.25.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.25.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.26.linear_attn.in_proj_a",
+    "language_model.model.layers.26.linear_attn.in_proj_b",
+    "language_model.model.layers.26.linear_attn.in_proj_qkv",
+    "language_model.model.layers.26.linear_attn.in_proj_z",
+    "language_model.model.layers.26.linear_attn.out_proj",
+    "language_model.model.layers.26.mlp.shared_expert.down_proj",
+    "language_model.model.layers.26.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.26.mlp.shared_expert.up_proj",
+    "language_model.model.layers.26.mlp.shared_expert_gate",
+    "language_model.model.layers.26.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.26.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.26.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.27.mlp.shared_expert.down_proj",
+    "language_model.model.layers.27.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.27.mlp.shared_expert.up_proj",
+    "language_model.model.layers.27.mlp.shared_expert_gate",
+    "language_model.model.layers.27.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.27.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.27.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.27.self_attn.k_proj",
+    "language_model.model.layers.27.self_attn.o_proj",
+    "language_model.model.layers.27.self_attn.q_proj",
+    "language_model.model.layers.27.self_attn.v_proj",
+    "language_model.model.layers.28.linear_attn.in_proj_a",
+    "language_model.model.layers.28.linear_attn.in_proj_b",
+    "language_model.model.layers.28.linear_attn.in_proj_qkv",
+    "language_model.model.layers.28.linear_attn.in_proj_z",
+    "language_model.model.layers.28.linear_attn.out_proj",
+    "language_model.model.layers.28.mlp.shared_expert.down_proj",
+    "language_model.model.layers.28.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.28.mlp.shared_expert.up_proj",
+    "language_model.model.layers.28.mlp.shared_expert_gate",
+    "language_model.model.layers.28.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.28.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.28.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.29.linear_attn.in_proj_a",
+    "language_model.model.layers.29.linear_attn.in_proj_b",
+    "language_model.model.layers.29.linear_attn.in_proj_qkv",
+    "language_model.model.layers.29.linear_attn.in_proj_z",
+    "language_model.model.layers.29.linear_attn.out_proj",
+    "language_model.model.layers.29.mlp.shared_expert.down_proj",
+    "language_model.model.layers.29.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.29.mlp.shared_expert.up_proj",
+    "language_model.model.layers.29.mlp.shared_expert_gate",
+    "language_model.model.layers.29.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.29.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.29.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.3.mlp.shared_expert.down_proj",
+    "language_model.model.layers.3.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.3.mlp.shared_expert.up_proj",
+    "language_model.model.layers.3.mlp.shared_expert_gate",
+    "language_model.model.layers.3.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.3.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.3.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.3.self_attn.k_proj",
+    "language_model.model.layers.3.self_attn.o_proj",
+    "language_model.model.layers.3.self_attn.q_proj",
+    "language_model.model.layers.3.self_attn.v_proj",
+    "language_model.model.layers.30.linear_attn.in_proj_a",
+    "language_model.model.layers.30.linear_attn.in_proj_b",
+    "language_model.model.layers.30.linear_attn.in_proj_qkv",
+    "language_model.model.layers.30.linear_attn.in_proj_z",
+    "language_model.model.layers.30.linear_attn.out_proj",
+    "language_model.model.layers.30.mlp.shared_expert.down_proj",
+    "language_model.model.layers.30.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.30.mlp.shared_expert.up_proj",
+    "language_model.model.layers.30.mlp.shared_expert_gate",
+    "language_model.model.layers.30.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.30.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.30.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.31.mlp.shared_expert.down_proj",
+    "language_model.model.layers.31.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.31.mlp.shared_expert.up_proj",
+    "language_model.model.layers.31.mlp.shared_expert_gate",
+    "language_model.model.layers.31.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.31.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.31.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.31.self_attn.k_proj",
+    "language_model.model.layers.31.self_attn.o_proj",
+    "language_model.model.layers.31.self_attn.q_proj",
+    "language_model.model.layers.31.self_attn.v_proj",
+    "language_model.model.layers.32.linear_attn.in_proj_a",
+    "language_model.model.layers.32.linear_attn.in_proj_b",
+    "language_model.model.layers.32.linear_attn.in_proj_qkv",
+    "language_model.model.layers.32.linear_attn.in_proj_z",
+    "language_model.model.layers.32.linear_attn.out_proj",
+    "language_model.model.layers.32.mlp.shared_expert.down_proj",
+    "language_model.model.layers.32.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.32.mlp.shared_expert.up_proj",
+    "language_model.model.layers.32.mlp.shared_expert_gate",
+    "language_model.model.layers.32.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.32.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.32.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.33.linear_attn.in_proj_a",
+    "language_model.model.layers.33.linear_attn.in_proj_b",
+    "language_model.model.layers.33.linear_attn.in_proj_qkv",
+    "language_model.model.layers.33.linear_attn.in_proj_z",
+    "language_model.model.layers.33.linear_attn.out_proj",
+    "language_model.model.layers.33.mlp.shared_expert.down_proj",
+    "language_model.model.layers.33.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.33.mlp.shared_expert.up_proj",
+    "language_model.model.layers.33.mlp.shared_expert_gate",
+    "language_model.model.layers.33.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.33.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.33.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.34.linear_attn.in_proj_a",
+    "language_model.model.layers.34.linear_attn.in_proj_b",
+    "language_model.model.layers.34.linear_attn.in_proj_qkv",
+    "language_model.model.layers.34.linear_attn.in_proj_z",
+    "language_model.model.layers.34.linear_attn.out_proj",
+    "language_model.model.layers.34.mlp.shared_expert.down_proj",
+    "language_model.model.layers.34.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.34.mlp.shared_expert.up_proj",
+    "language_model.model.layers.34.mlp.shared_expert_gate",
+    "language_model.model.layers.34.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.34.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.34.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.35.mlp.shared_expert.down_proj",
+    "language_model.model.layers.35.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.35.mlp.shared_expert.up_proj",
+    "language_model.model.layers.35.mlp.shared_expert_gate",
+    "language_model.model.layers.35.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.35.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.35.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.35.self_attn.k_proj",
+    "language_model.model.layers.35.self_attn.o_proj",
+    "language_model.model.layers.35.self_attn.q_proj",
+    "language_model.model.layers.35.self_attn.v_proj",
+    "language_model.model.layers.36.linear_attn.in_proj_a",
+    "language_model.model.layers.36.linear_attn.in_proj_b",
+    "language_model.model.layers.36.linear_attn.in_proj_qkv",
+    "language_model.model.layers.36.linear_attn.in_proj_z",
+    "language_model.model.layers.36.linear_attn.out_proj",
+    "language_model.model.layers.36.mlp.shared_expert.down_proj",
+    "language_model.model.layers.36.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.36.mlp.shared_expert.up_proj",
+    "language_model.model.layers.36.mlp.shared_expert_gate",
+    "language_model.model.layers.36.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.36.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.36.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.37.linear_attn.in_proj_a",
+    "language_model.model.layers.37.linear_attn.in_proj_b",
+    "language_model.model.layers.37.linear_attn.in_proj_qkv",
+    "language_model.model.layers.37.linear_attn.in_proj_z",
+    "language_model.model.layers.37.linear_attn.out_proj",
+    "language_model.model.layers.37.mlp.shared_expert.down_proj",
+    "language_model.model.layers.37.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.37.mlp.shared_expert.up_proj",
+    "language_model.model.layers.37.mlp.shared_expert_gate",
+    "language_model.model.layers.37.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.37.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.37.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.38.linear_attn.in_proj_a",
+    "language_model.model.layers.38.linear_attn.in_proj_b",
+    "language_model.model.layers.38.linear_attn.in_proj_qkv",
+    "language_model.model.layers.38.linear_attn.in_proj_z",
+    "language_model.model.layers.38.linear_attn.out_proj",
+    "language_model.model.layers.38.mlp.shared_expert.down_proj",
+    "language_model.model.layers.38.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.38.mlp.shared_expert.up_proj",
+    "language_model.model.layers.38.mlp.shared_expert_gate",
+    "language_model.model.layers.38.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.38.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.38.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.39.mlp.shared_expert.down_proj",
+    "language_model.model.layers.39.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.39.mlp.shared_expert.up_proj",
+    "language_model.model.layers.39.mlp.shared_expert_gate",
+    "language_model.model.layers.39.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.39.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.39.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.39.self_attn.k_proj",
+    "language_model.model.layers.39.self_attn.o_proj",
+    "language_model.model.layers.39.self_attn.q_proj",
+    "language_model.model.layers.39.self_attn.v_proj",
+    "language_model.model.layers.4.linear_attn.in_proj_a",
+    "language_model.model.layers.4.linear_attn.in_proj_b",
+    "language_model.model.layers.4.linear_attn.in_proj_qkv",
+    "language_model.model.layers.4.linear_attn.in_proj_z",
+    "language_model.model.layers.4.linear_attn.out_proj",
+    "language_model.model.layers.4.mlp.shared_expert.down_proj",
+    "language_model.model.layers.4.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.4.mlp.shared_expert.up_proj",
+    "language_model.model.layers.4.mlp.shared_expert_gate",
+    "language_model.model.layers.4.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.4.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.4.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.5.linear_attn.in_proj_a",
+    "language_model.model.layers.5.linear_attn.in_proj_b",
+    "language_model.model.layers.5.linear_attn.in_proj_qkv",
+    "language_model.model.layers.5.linear_attn.in_proj_z",
+    "language_model.model.layers.5.linear_attn.out_proj",
+    "language_model.model.layers.5.mlp.shared_expert.down_proj",
+    "language_model.model.layers.5.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.5.mlp.shared_expert.up_proj",
+    "language_model.model.layers.5.mlp.shared_expert_gate",
+    "language_model.model.layers.5.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.5.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.5.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.6.linear_attn.in_proj_a",
+    "language_model.model.layers.6.linear_attn.in_proj_b",
+    "language_model.model.layers.6.linear_attn.in_proj_qkv",
+    "language_model.model.layers.6.linear_attn.in_proj_z",
+    "language_model.model.layers.6.linear_attn.out_proj",
+    "language_model.model.layers.6.mlp.shared_expert.down_proj",
+    "language_model.model.layers.6.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.6.mlp.shared_expert.up_proj",
+    "language_model.model.layers.6.mlp.shared_expert_gate",
+    "language_model.model.layers.6.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.6.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.6.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.7.mlp.shared_expert.down_proj",
+    "language_model.model.layers.7.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.7.mlp.shared_expert.up_proj",
+    "language_model.model.layers.7.mlp.shared_expert_gate",
+    "language_model.model.layers.7.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.7.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.7.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.7.self_attn.k_proj",
+    "language_model.model.layers.7.self_attn.o_proj",
+    "language_model.model.layers.7.self_attn.q_proj",
+    "language_model.model.layers.7.self_attn.v_proj",
+    "language_model.model.layers.8.linear_attn.in_proj_a",
+    "language_model.model.layers.8.linear_attn.in_proj_b",
+    "language_model.model.layers.8.linear_attn.in_proj_qkv",
+    "language_model.model.layers.8.linear_attn.in_proj_z",
+    "language_model.model.layers.8.linear_attn.out_proj",
+    "language_model.model.layers.8.mlp.shared_expert.down_proj",
+    "language_model.model.layers.8.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.8.mlp.shared_expert.up_proj",
+    "language_model.model.layers.8.mlp.shared_expert_gate",
+    "language_model.model.layers.8.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.8.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.8.mlp.switch_mlp.up_proj",
+    "language_model.model.layers.9.linear_attn.in_proj_a",
+    "language_model.model.layers.9.linear_attn.in_proj_b",
+    "language_model.model.layers.9.linear_attn.in_proj_qkv",
+    "language_model.model.layers.9.linear_attn.in_proj_z",
+    "language_model.model.layers.9.linear_attn.out_proj",
+    "language_model.model.layers.9.mlp.shared_expert.down_proj",
+    "language_model.model.layers.9.mlp.shared_expert.gate_proj",
+    "language_model.model.layers.9.mlp.shared_expert.up_proj",
+    "language_model.model.layers.9.mlp.shared_expert_gate",
+    "language_model.model.layers.9.mlp.switch_mlp.down_proj",
+    "language_model.model.layers.9.mlp.switch_mlp.gate_proj",
+    "language_model.model.layers.9.mlp.switch_mlp.up_proj",
+    "language_model.mtp.layers.0.mlp.shared_expert.down_proj",
+    "language_model.mtp.layers.0.mlp.shared_expert.gate_proj",
+    "language_model.mtp.layers.0.mlp.shared_expert.up_proj",
+    "language_model.mtp.layers.0.mlp.shared_expert_gate",
+    "language_model.mtp.layers.0.mlp.switch_mlp.down_proj",
+    "language_model.mtp.layers.0.mlp.switch_mlp.gate_proj",
+    "language_model.mtp.layers.0.mlp.switch_mlp.up_proj",
+    "language_model.mtp.layers.0.self_attn.k_proj",
+    "language_model.mtp.layers.0.self_attn.o_proj",
+    "language_model.mtp.layers.0.self_attn.q_proj",
+    "language_model.mtp.layers.0.self_attn.v_proj"
+  ],
+  "missing": [
+    "language_model.lm_head",
+    "language_model.model.embed_tokens"
+  ],
+  "mismatched": [],
+  "zero_count_experts": 0
+}

preprocessor_config.json

@@ -0,0 +1,21 @@
+{
+    "size": {
+        "longest_edge": 16777216,
+        "shortest_edge": 65536
+    },
+    "patch_size": 16,
+    "temporal_patch_size": 2,
+    "merge_size": 2,
+    "image_mean": [
+        0.5,
+        0.5,
+        0.5
+    ],
+    "image_std": [
+        0.5,
+        0.5,
+        0.5
+    ],
+    "processor_class": "Qwen3VLProcessor",
+    "image_processor_type": "Qwen2VLImageProcessorFast"
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5f9e4d4901a92b997e463c1f46055088b6cca5ca61a6522d1b9f64c4bb81cb42
+size 12807982

tokenizer_config.json

@@ -0,0 +1,305 @@
+{
+    "add_prefix_space": false,
+    "added_tokens_decoder": {
+        "248044": {
+            "content": "<|endoftext|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248045": {
+            "content": "<|im_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248046": {
+            "content": "<|im_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248047": {
+            "content": "<|object_ref_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248048": {
+            "content": "<|object_ref_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248049": {
+            "content": "<|box_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248050": {
+            "content": "<|box_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248051": {
+            "content": "<|quad_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248052": {
+            "content": "<|quad_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248053": {
+            "content": "<|vision_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248054": {
+            "content": "<|vision_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248055": {
+            "content": "<|vision_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248056": {
+            "content": "<|image_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248057": {
+            "content": "<|video_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248058": {
+            "content": "<tool_call>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248059": {
+            "content": "</tool_call>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248060": {
+            "content": "<|fim_prefix|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248061": {
+            "content": "<|fim_middle|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248062": {
+            "content": "<|fim_suffix|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248063": {
+            "content": "<|fim_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248064": {
+            "content": "<|repo_name|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248065": {
+            "content": "<|file_sep|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248066": {
+            "content": "<tool_response>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248067": {
+            "content": "</tool_response>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248068": {
+            "content": "<think>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248069": {
+            "content": "</think>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248070": {
+            "content": "<|audio_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248071": {
+            "content": "<|audio_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248072": {
+            "content": "<tts_pad>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248073": {
+            "content": "<tts_text_bos>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248074": {
+            "content": "<tts_text_eod>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248075": {
+            "content": "<tts_text_bos_single>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248076": {
+            "content": "<|audio_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        }
+    },
+    "additional_special_tokens": [
+        "<|im_start|>",
+        "<|im_end|>",
+        "<|object_ref_start|>",
+        "<|object_ref_end|>",
+        "<|box_start|>",
+        "<|box_end|>",
+        "<|quad_start|>",
+        "<|quad_end|>",
+        "<|vision_start|>",
+        "<|vision_end|>",
+        "<|vision_pad|>",
+        "<|image_pad|>",
+        "<|video_pad|>"
+    ],
+    "bos_token": null,
+    "chat_template": "{%- set image_count = namespace(value=0) %}\n{%- set video_count = namespace(value=0) %}\n{%- macro render_content(content, do_vision_count, is_system_content=false) %}\n    {%- if content is string %}\n        {{- content }}\n    {%- elif content is iterable and content is not mapping %}\n        {%- for item in content %}\n            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}\n                {%- if is_system_content %}\n                    {{- raise_exception('System message cannot contain images.') }}\n                {%- endif %}\n                {%- if do_vision_count %}\n                    {%- set image_count.value = image_count.value + 1 %}\n                {%- endif %}\n                {%- if add_vision_id %}\n                    {{- 'Picture ' ~ image_count.value ~ ': ' }}\n                {%- endif %}\n                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}\n            {%- elif 'video' in item or item.type == 'video' %}\n                {%- if is_system_content %}\n                    {{- raise_exception('System message cannot contain videos.') }}\n                {%- endif %}\n                {%- if do_vision_count %}\n                    {%- set video_count.value = video_count.value + 1 %}\n                {%- endif %}\n                {%- if add_vision_id %}\n                    {{- 'Video ' ~ video_count.value ~ ': ' }}\n                {%- endif %}\n                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}\n            {%- elif 'text' in item %}\n                {{- item.text }}\n            {%- else %}\n                {{- raise_exception('Unexpected item type in content.') }}\n            {%- endif %}\n        {%- endfor %}\n    {%- elif content is none or content is undefined %}\n        {{- '' }}\n    {%- else %}\n        {{- raise_exception('Unexpected content type.') }}\n    {%- endif %}\n{%- endmacro %}\n{%- if not messages %}\n    {{- raise_exception('No messages provided.') }}\n{%- endif %}\n{%- if tools and tools is iterable and tools is not mapping %}\n    {{- '<|im_start|>system\\n' }}\n    {{- \"# Tools\\n\\nYou have access to the following functions:\\n\\n<tools>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\n</tools>\" }}\n    {{- '\\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>' }}\n    {%- if messages[0].role == 'system' %}\n        {%- set content = render_content(messages[0].content, false, true)|trim %}\n        {%- if content %}\n            {{- '\\n\\n' + content }}\n        {%- endif %}\n    {%- endif %}\n    {{- '<|im_end|>\\n' }}\n{%- else %}\n    {%- if messages[0].role == 'system' %}\n        {%- set content = render_content(messages[0].content, false, true)|trim %}\n        {{- '<|im_start|>system\\n' + content + '<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}\n{%- for message in messages[::-1] %}\n    {%- set index = (messages|length - 1) - loop.index0 %}\n    {%- if ns.multi_step_tool and message.role == \"user\" %}\n        {%- set content = render_content(message.content, false)|trim %}\n        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}\n            {%- set ns.multi_step_tool = false %}\n            {%- set ns.last_query_index = index %}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if ns.multi_step_tool %}\n    {{- raise_exception('No user query found in messages.') }}\n{%- endif %}\n{%- for message in messages %}\n    {%- set content = render_content(message.content, true)|trim %}\n    {%- if message.role == \"system\" %}\n        {%- if not loop.first %}\n            {{- raise_exception('System message must be at the beginning.') }}\n        {%- endif %}\n    {%- elif message.role == \"user\" %}\n        {{- '<|im_start|>' + message.role + '\\n' + content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {%- set reasoning_content = '' %}\n        {%- if message.reasoning_content is string %}\n            {%- set reasoning_content = message.reasoning_content %}\n        {%- else %}\n            {%- if '</think>' in content %}\n                {%- set reasoning_content = content.split('</think>')[0].rstrip('\\n').split('<think>')[-1].lstrip('\\n') %}\n                {%- set content = content.split('</think>')[-1].lstrip('\\n') %}\n            {%- endif %}\n        {%- endif %}\n        {%- set reasoning_content = reasoning_content|trim %}\n        {%- if (preserve_thinking is defined and preserve_thinking is true) or (loop.index0 > ns.last_query_index) %}\n            {{- '<|im_start|>' + message.role + '\\n<think>\\n' + reasoning_content + '\\n</think>\\n\\n' + content }}\n        {%- else %}\n            {{- '<|im_start|>' + message.role + '\\n' + content }}\n        {%- endif %}\n        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}\n            {%- for tool_call in message.tool_calls %}\n                {%- if tool_call.function is defined %}\n                    {%- set tool_call = tool_call.function %}\n                {%- endif %}\n                {%- if loop.first %}\n                    {%- if content|trim %}\n                        {{- '\\n\\n<tool_call>\\n<function=' + tool_call.name + '>\\n' }}\n                    {%- else %}\n                        {{- '<tool_call>\\n<function=' + tool_call.name + '>\\n' }}\n                    {%- endif %}\n                {%- else %}\n                    {{- '\\n<tool_call>\\n<function=' + tool_call.name + '>\\n' }}\n                {%- endif %}\n                {%- if tool_call.arguments is defined %}\n                    {%- for args_name, args_value in tool_call.arguments|items %}\n                        {{- '<parameter=' + args_name + '>\\n' }}\n                        {%- set args_value = args_value | string if args_value is string else args_value | tojson | safe %}\n                        {{- args_value }}\n                        {{- '\\n</parameter>\\n' }}\n                    {%- endfor %}\n                {%- endif %}\n                {{- '</function>\\n</tool_call>' }}\n            {%- endfor %}\n        {%- endif %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if loop.previtem and loop.previtem.role != \"tool\" %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- content }}\n        {{- '\\n</tool_response>' }}\n        {%- if not loop.last and loop.nextitem.role != \"tool\" %}\n            {{- '<|im_end|>\\n' }}\n        {%- elif loop.last %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- else %}\n        {{- raise_exception('Unexpected message role.') }}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n    {%- if enable_thinking is defined and enable_thinking is false %}\n        {{- '<think>\\n\\n</think>\\n\\n' }}\n    {%- else %}\n        {{- '<think>\\n' }}\n    {%- endif %}\n{%- endif %}",
+    "clean_up_tokenization_spaces": false,
+    "eos_token": "<|im_end|>",
+    "errors": "replace",
+    "model_max_length": 262144,
+    "pad_token": "<|endoftext|>",
+    "split_special_tokens": false,
+    "tokenizer_class": "Qwen2Tokenizer",
+    "unk_token": null,
+    "add_bos_token": false,
+    "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+",
+    "extra_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|>"
+    }
+}