Add files using upload-large-folder tool

.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

LICENSE

@@ -0,0 +1,18 @@
+Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International
+
+Copyright © 2026 Gowtham Sridhar
+
+This license applies only to rights held by Gowtham Sridhar. Third-party and upstream materials remain subject to their applicable licenses and notices.
+
+This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
+
+You are free to share — copy and redistribute the material in any medium or format — under the following terms:
+
+- Attribution — You must give appropriate credit, provide a link to the license, and indicate if changes were made.
+- NonCommercial — You may not use the material for commercial purposes.
+- NoDerivatives — If you remix, transform, or build upon the material, you may not distribute the modified material.
+- No additional restrictions — You may not apply legal terms or technological measures that legally restrict others from doing anything the license permits.
+
+The complete legal code is available at:
+
+https://creativecommons.org/licenses/by-nc-nd/4.0/legalcode

README.md

@@ -0,0 +1,273 @@
+---
+license: cc-by-nc-nd-4.0
+base_model: Qwen/Qwen3.8-27B
+library_name: transformers
+pipeline_tag: image-text-to-text
+tags:
+  - qwen3.8
+  - agentic
+  - vision-language
+  - tool-calling
+  - quantized
+  - nvfp4
+---
+
+# ✦ qwen3.8-27b-agentic-nvfp4
+
+<p align="center">
+  <strong>Published and maintained by <a href="https://www.gowthamsridhar.com/">Gowtham Sridhar</a></strong>
+</p>
+
+An NVFP4 checkpoint of [Qwen3.8-27B](https://huggingface.co/Qwen/Qwen3.8-27B), prepared for agentic and multimodal applications. It combines native text, image, and video understanding with configurable reasoning and structured tool calling.
+
+> **Quick links:** [Base model](https://huggingface.co/Qwen/Qwen3.8-27B) · [DSpark companion](https://huggingface.co/gittensor-model-hub/Qwen3.8-27B-DSpark-NVFP4) · [vLLM documentation](https://docs.vllm.ai/en/latest/) · [SGLang documentation](https://docs.sglang.ai/)
+
+## ✨ What this model is for
+
+| Capability | Description |
+| --- | --- |
+| Agentic workflows | Tool use, multi-step tasks, and structured responses |
+| Multimodal input | Text, images, and video |
+| Reasoning control | Thinking can be enabled, preserved, or adjusted per request |
+| Deployment | Pre-quantized NVFP4 weights for compatible serving runtimes |
+
+The model supports a native context window of up to 262,144 tokens. Test the context length, memory settings, and tool-call parser in your own deployment before production use.
+
+## 🧭 Before you start
+
+Set the model identifier once. Replace the placeholder with the Hugging Face repository name after publishing, or use a local path.
+
+```bash
+export MODEL_ID="<your-hugging-face-namespace>/qwen3.8-27b-agentic-nvfp4"
+# or: export MODEL_ID="/path/to/qwen3.8-27b-agentic-nvfp4"
+```
+
+This checkpoint contains ModelOpt NVFP4 metadata. Use a current runtime that supports the model architecture and this quantization format. Do not pass a second quantization option when loading this already-quantized checkpoint.
+
+## 🚀 Serve with vLLM
+
+Start with standard serving:
+
+```bash
+vllm serve "$MODEL_ID" \
+  --trust-remote-code \
+  --kv-cache-dtype fp8 \
+  --max-model-len 262144 \
+  --reasoning-parser qwen3 \
+  --enable-auto-tool-choice \
+  --tool-call-parser qwen3_xml
+```
+
+The template emits XML tool calls by default. If your installed vLLM release does not provide the `qwen3_xml` parser, omit the last two tool-choice flags and handle tool calls in your application.
+
+### Choose one serving mode
+
+| Mode | When to use it | Server option |
+| --- | --- | --- |
+| Standard | First run, long-context work, or highest compatibility | No speculative option |
+| Native MTP | The installed vLLM release recognizes this model's MTP head | `--speculative-config '{"method":"mtp",...}'` |
+| DSpark | Using the paired external draft model in an SGLang build with DSpark support | `--speculative-algorithm DSPARK` |
+
+Start with standard serving and confirm a normal chat request first. Then enable exactly one speculative mode. Do not combine native MTP and DSpark in the same server.
+
+### 💬 OpenAI-compatible request
+
+After the server starts, use the OpenAI-compatible endpoint:
+
+```python
+from openai import OpenAI
+
+client = OpenAI(base_url="http://localhost:8000/v1", api_key="not-needed")
+
+response = client.chat.completions.create(
+    model="qwen3.8-27b-agentic-nvfp4",
+    messages=[
+        {"role": "user", "content": "Outline a practical plan for organizing a research project."}
+    ],
+    temperature=0.7,
+)
+
+print(response.choices[0].message.content)
+```
+
+### 🧠 Reasoning and direct-response modes
+
+Pass chat-template controls in `extra_body`. The following request keeps the default high reasoning effort:
+
+```python
+response = client.chat.completions.create(
+    model="qwen3.8-27b-agentic-nvfp4",
+    messages=[{"role": "user", "content": "Compare two project plans and recommend one."}],
+    extra_body={
+        "chat_template_kwargs": {
+            "enable_thinking": True,
+            "reasoning_effort": "xhigh",
+            "preserve_thinking": True,
+        }
+    },
+)
+```
+
+For a concise direct response, disable thinking explicitly:
+
+```python
+extra_body={"chat_template_kwargs": {"enable_thinking": False}}
+```
+
+### 🛠️ Tool calling
+
+The standard vLLM command above enables automatic tool selection when the `qwen3_xml` parser is available. Supply tools using the OpenAI-compatible schema:
+
+```python
+tools = [
+    {
+        "type": "function",
+        "function": {
+            "name": "get_weather",
+            "description": "Get the current weather for a city.",
+            "parameters": {
+                "type": "object",
+                "properties": {"city": {"type": "string"}},
+                "required": ["city"],
+            },
+        },
+    }
+]
+
+response = client.chat.completions.create(
+    model="qwen3.8-27b-agentic-nvfp4",
+    messages=[{"role": "user", "content": "What is the weather in Vienna?"}],
+    tools=tools,
+    tool_choice="auto",
+    extra_body={
+        "chat_template_kwargs": {
+            "enable_thinking": False,
+            "auto_disable_thinking_with_tools": True,
+        }
+    },
+)
+
+print(response.choices[0].message.tool_calls)
+```
+
+Execute returned tools in your application, append their results as `tool` messages, then send the updated conversation back to the model. The template accepts tool arguments provided either as an object or as a JSON string.
+
+### 🖼️ Image input
+
+The OpenAI-compatible endpoint accepts image content alongside text. The server applies the model's chat template and image placeholders automatically:
+
+```python
+response = client.chat.completions.create(
+    model="qwen3.8-27b-agentic-nvfp4",
+    messages=[
+        {
+            "role": "user",
+            "content": [
+                {"type": "text", "text": "Describe this image and list the main objects."},
+                {"type": "image_url", "image_url": {"url": "https://example.com/image.jpg"}},
+            ],
+        }
+    ],
+)
+```
+
+## ⚡ Native MTP with vLLM
+
+This checkpoint includes a native multi-token prediction (MTP) head. MTP lets the model propose a small number of tokens before the target model verifies them. Start with one speculative token, validate your workload, and increase only if your environment benefits.
+
+```bash
+vllm serve "$MODEL_ID" \
+  --trust-remote-code \
+  --kv-cache-dtype fp8 \
+  --max-model-len 262144 \
+  --speculative-config '{"method":"mtp","num_speculative_tokens":1}'
+```
+
+`num_speculative_tokens` is the speculative depth. Keep it at `1` for the first validation run. If the server starts and your output checks pass, tune one setting at a time and retain the configuration that works best for your workload.
+
+Use either native MTP or the external DSpark draft model below—not both in the same server. vLLM requires MTP support for the model family in the installed release; if startup rejects the MTP configuration, update to a release that supports this architecture or use standard serving.
+
+## 🔗 Serve with SGLang
+
+For normal SGLang serving, launch the model first and use the OpenAI-compatible endpoint on port `30000`:
+
+```bash
+python3 -m sglang.launch_server \
+  --model-path "$MODEL_ID" \
+  --trust-remote-code \
+  --port 30000
+```
+
+For automatic tool-call parsing in SGLang, add `--tool-call-parser qwen3_coder` when that parser is available in your installed build. Otherwise, keep the server unparsed and process the template's XML tool-call blocks in your application.
+
+## 🔭 DSpark companion model
+
+For draft-model speculative decoding, use [Qwen3.8-27B-DSpark-NVFP4](https://huggingface.co/gittensor-model-hub/Qwen3.8-27B-DSpark-NVFP4) as a separate companion checkpoint. DSpark is an alternative to the embedded MTP head: it drafts candidate tokens and the target model verifies them.
+
+### DSpark with SGLang
+
+DSpark support is actively evolving in SGLang. Use an SGLang build that exposes the `DSPARK` speculative algorithm, then launch the target and draft model together:
+
+```bash
+export DSPARK_MODEL_ID="gittensor-model-hub/Qwen3.8-27B-DSpark-NVFP4"
+
+python -m sglang.launch_server \
+  --model-path "$MODEL_ID" \
+  --trust-remote-code \
+  --speculative-algorithm DSPARK \
+  --speculative-draft-model-path "$DSPARK_MODEL_ID" \
+  --port 30000
+```
+
+SGLang reads the DSpark draft configuration to determine its verification window. Keep the target and draft model versions paired. If the server asks for a draft block size, use the value stored in the DSpark checkpoint rather than an arbitrary override.
+
+## 💡 Chat template
+
+This chat template has been improved for reliable agentic and tool-calling workflows.
+
+| Request option | Default | Purpose |
+| --- | --- | --- |
+| `reasoning_effort` | `xhigh` | Choose `xhigh`, `medium`, or `low` reasoning depth |
+| `enable_thinking` | `true` | Enable or disable thinking output |
+| `preserve_thinking` | `true` | Keep prior thinking in conversation history |
+| `tool_call_format` | `xml` | Use `xml` or `json` tool-call output |
+| `auto_disable_thinking_with_tools` | `false` | Turn off thinking automatically when tools are present |
+| `continue_final_message` | `false` | Continue an assistant message instead of adding a new turn |
+
+## 🧩 Transformers
+
+```python
+from transformers import AutoModelForImageTextToText, AutoProcessor
+
+model_id = "<your-hugging-face-namespace>/qwen3.8-27b-agentic-nvfp4"
+
+model = AutoModelForImageTextToText.from_pretrained(
+    model_id,
+    trust_remote_code=True,
+)
+processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)
+```
+
+Replace `model_id` with the published repository name or local path.
+
+## 📌 Model details
+
+| Item | Value |
+| --- | --- |
+| Base model | [Qwen3.8-27B](https://huggingface.co/Qwen/Qwen3.8-27B) |
+| Architecture | Vision-language, image-text-to-text |
+| Quantization | NVFP4 |
+| Context window | Up to 262,144 tokens |
+| Intended use | Agentic, multimodal, and general-purpose generation |
+
+## ✓ Quality and responsible use
+
+This release is intended to retain the base model's general-purpose, multimodal, and agentic capabilities. Evaluate quality, tool reliability, safety, and resource use on your own tasks before production deployment.
+
+## Attribution
+
+This is a derived checkpoint based on Qwen3.8-27B.
+
+## License
+
+This repository is licensed under [CC BY-NC-ND 4.0](LICENSE). You may share it with attribution for non-commercial purposes. You may not distribute modified versions.

chat_template.jinja

@@ -0,0 +1,284 @@
+{%- set template_version = "qwen3.8-gittensor-safe-v2" -%}
+{%- set tool_format = tool_call_format if tool_call_format is defined else "xml" -%}
+{%- set add_vision_id = add_vision_id if add_vision_id is defined else false -%}
+{%- set enable_thinking = enable_thinking if enable_thinking is defined else true -%}
+{%- set preserve_thinking = preserve_thinking if preserve_thinking is defined else true -%}
+{%- set auto_disable_thinking_with_tools = auto_disable_thinking_with_tools if auto_disable_thinking_with_tools is defined else false -%}
+{%- set max_tool_arg_chars = max_tool_arg_chars if max_tool_arg_chars is defined else 0 -%}
+{%- set max_tool_response_chars = max_tool_response_chars if max_tool_response_chars is defined else 0 -%}
+{%- set has_tools = tools is defined and tools and tools is iterable and tools is not mapping -%}
+{%- set image_counter = namespace(value=0) -%}
+{%- set video_counter = namespace(value=0) -%}
+{%- set state = namespace(thinking=enable_thinking, previous_role="") -%}
+{%- if auto_disable_thinking_with_tools and has_tools -%}
+  {%- set state.thinking = false -%}
+{%- endif -%}
+
+{%- macro render_content(value, count_vision=false, system_content=false) -%}
+  {%- if value is string -%}
+    {{- value -}}
+  {%- elif value is iterable and value is not mapping -%}
+    {%- for part in value -%}
+      {%- if part is mapping -%}
+        {%- set part_type = part.type if part.type is defined else "" -%}
+        {%- if part_type == "image" or "image" in part or "image_url" in part -%}
+          {%- if system_content -%}
+            {{- raise_exception("System/developer messages cannot contain images.") -}}
+          {%- endif -%}
+          {%- if count_vision -%}
+            {%- set image_counter.value = image_counter.value + 1 -%}
+          {%- endif -%}
+          {%- if add_vision_id -%}
+            {{- "Picture " ~ image_counter.value ~ ": " -}}
+          {%- endif -%}
+          {{- "<|vision_start|><|image_pad|><|vision_end|>" -}}
+        {%- elif part_type == "video" or "video" in part -%}
+          {%- if system_content -%}
+            {{- raise_exception("System/developer messages cannot contain videos.") -}}
+          {%- endif -%}
+          {%- if count_vision -%}
+            {%- set video_counter.value = video_counter.value + 1 -%}
+          {%- endif -%}
+          {%- if add_vision_id -%}
+            {{- "Video " ~ video_counter.value ~ ": " -}}
+          {%- endif -%}
+          {{- "<|vision_start|><|video_pad|><|vision_end|>" -}}
+        {%- elif "text" in part -%}
+          {{- part.text -}}
+        {%- else -%}
+          {{- raise_exception("Unsupported multimodal content item.") -}}
+        {%- endif -%}
+      {%- else -%}
+        {{- part | string -}}
+      {%- endif -%}
+    {%- endfor -%}
+  {%- elif value is none or value is undefined -%}
+    {{- "" -}}
+  {%- else -%}
+    {{- raise_exception("Unsupported message content type.") -}}
+  {%- endif -%}
+{%- endmacro -%}
+
+{%- if not messages -%}
+  {{- raise_exception("No messages provided.") -}}
+{%- endif -%}
+
+{# Pull only the leading system/developer message into the tool system block. #}
+{%- set first_role = messages[0].role -%}
+{%- if first_role == "system" or first_role == "developer" -%}
+  {%- set leading_system = messages[0] -%}
+  {%- set conversation = messages[1:] -%}
+{%- else -%}
+  {%- set leading_system = none -%}
+  {%- set conversation = messages -%}
+{%- endif -%}
+
+{%- set system_text = "" -%}
+{%- if leading_system is not none -%}
+  {%- set system_text = render_content(leading_system.content, false, true) | trim -%}
+  {%- if "<|think_off|>" in system_text -%}
+    {%- set state.thinking = false -%}
+    {%- set system_text = system_text.split("<|think_off|>") | join("") | trim -%}
+  {%- elif "<|think_on|>" in system_text -%}
+    {%- set state.thinking = true -%}
+    {%- set system_text = system_text.split("<|think_on|>") | join("") | trim -%}
+  {%- endif -%}
+{%- endif -%}
+
+{# Reasoning-effort steering. Native Qwen3.8 behaviour: xhigh (default), medium, low.
+   medium intentionally emits no instructions. Honours the resolved thinking state, so
+   enable_thinking=false, <|think_off|> and auto_disable_thinking_with_tools all suppress it. #}
+{%- set reasoning_instructions = "" -%}
+{%- if state.thinking -%}
+  {%- set resolved_reasoning_effort = reasoning_effort if (reasoning_effort is defined and reasoning_effort) else "xhigh" -%}
+  {%- if resolved_reasoning_effort not in ("xhigh", "medium", "low") -%}
+    {{- raise_exception("Unexpected reasoning effort " ~ resolved_reasoning_effort ~ ". Supported types are xhigh (default), medium, and low.") -}}
+  {%- endif -%}
+  {%- if resolved_reasoning_effort == "xhigh" -%}
+    {%- set reasoning_instructions = "Reasoning effort is set to xhigh. Please think carefully through the task, validate key assumptions, consider plausible alternatives, and prioritize correctness, consistency, and clarity in the final answer." -%}
+  {%- elif resolved_reasoning_effort == "low" -%}
+    {%- set reasoning_instructions = "Reasoning effort is set to low. Keep your thinking brief and focused, moving directly to the conclusion without unnecessary elaboration." -%}
+  {%- endif -%}
+{%- endif -%}
+
+{%- if has_tools -%}
+  {{- "<|im_start|>system\n" -}}
+  {%- if reasoning_instructions -%}
+    {{- reasoning_instructions ~ "\n\n" -}}
+  {%- endif -%}
+  {{- "# Tools\n\nYou have access to the following functions.\n\n<tools>" -}}
+  {%- for tool in tools -%}
+    {{- "\n" ~ (tool | tojson) -}}
+  {%- endfor -%}
+  {{- "\n</tools>\n\n" -}}
+  {%- if tool_format == "json" -%}
+    {{- "When a tool is needed, emit one or more tool calls in this exact structure:\n<tool_call>\n{\"name\": \"function_name\", \"arguments\": {\"parameter\": \"value\"}}\n</tool_call>\n" -}}
+  {%- else -%}
+    {{- "When a tool is needed, emit one or more tool calls in this exact structure:\n<tool_call>\n<function=function_name>\n<parameter=parameter_name>\nvalue\n</parameter>\n</function>\n</tool_call>\n" -}}
+  {%- endif -%}
+  {{- "If you call a tool, output only an optional <think>...</think> block followed immediately by the <tool_call> block(s); do not add ordinary assistant text before or after the calls. For multiple calls, emit separate fully closed <tool_call> blocks. If no tool is needed, answer normally without a tool call." -}}
+  {%- if system_text -%}
+    {{- "\n\n" ~ system_text -}}
+  {%- endif -%}
+  {{- "<|im_end|>\n" -}}
+{%- elif system_text or reasoning_instructions -%}
+  {{- "<|im_start|>system\n" -}}
+  {%- if reasoning_instructions -%}
+    {{- reasoning_instructions -}}
+    {%- if system_text -%}
+      {{- "\n\n" -}}
+    {%- endif -%}
+  {%- endif -%}
+  {{- system_text ~ "<|im_end|>\n" -}}
+{%- endif -%}
+
+{# Track the newest real user query. Tool responses are role=tool, so no heuristic is needed. #}
+{%- set query_state = namespace(last_user_index=-1) -%}
+{%- for item in conversation -%}
+  {%- if item.role == "user" -%}
+    {%- set candidate_user = render_content(item.content, false) | trim -%}
+    {%- if not (candidate_user.startswith("<tool_response>") and candidate_user.endswith("</tool_response>")) -%}
+      {%- set query_state.last_user_index = loop.index0 -%}
+    {%- endif -%}
+  {%- endif -%}
+{%- endfor -%}
+
+{%- for message in conversation -%}
+  {%- set role = message.role -%}
+  {%- set system_like = role == "system" or role == "developer" -%}
+  {%- set content = render_content(message.content, true, system_like) | trim -%}
+
+  {%- set wrapped_tool_response = role == "user" and content.startswith("<tool_response>") and content.endswith("</tool_response>") -%}
+  {%- if system_like or (role == "user" and not wrapped_tool_response) -%}
+    {%- if "<|think_off|>" in content -%}
+      {%- set state.thinking = false -%}
+      {%- set content = content.split("<|think_off|>") | join("") | trim -%}
+    {%- elif "<|think_on|>" in content -%}
+      {%- set state.thinking = true -%}
+      {%- set content = content.split("<|think_on|>") | join("") | trim -%}
+    {%- endif -%}
+  {%- endif -%}
+
+  {%- if system_like -%}
+    {{- "<|im_start|>system\n" ~ content ~ "<|im_end|>\n" -}}
+
+  {%- elif role == "user" -%}
+    {{- "<|im_start|>user\n" ~ content ~ "<|im_end|>\n" -}}
+
+  {%- elif role == "assistant" -%}
+    {%- set reasoning = "" -%}
+    {%- if message.reasoning_content is defined and message.reasoning_content is not none -%}
+      {%- set reasoning = message.reasoning_content if message.reasoning_content is string else (message.reasoning_content | string) -%}
+    {%- elif message.thinking is defined and message.thinking is not none -%}
+      {%- set reasoning = message.thinking if message.thinking is string else (message.thinking | string) -%}
+    {%- else -%}
+      {%- set think_close = "" -%}
+      {%- set think_open = "<think>" -%}
+      {%- if content.startswith("</think>") -%}
+        {%- set think_close = "</think>" -%}
+      {%- elif content.startswith("</thinking>") -%}
+        {%- set think_close = "</thinking>" -%}
+        {%- set think_open = "<thinking>" -%}
+      {%- elif "\n</think>" in content -%}
+        {%- set think_close = "\n</think>" -%}
+      {%- elif "\n</thinking>" in content -%}
+        {%- set think_close = "\n</thinking>" -%}
+        {%- set think_open = "<thinking>" -%}
+      {%- elif "\n</ think>" in content -%}
+        {%- set think_close = "\n</ think>" -%}
+      {%- elif "\n</think >" in content -%}
+        {%- set think_close = "\n</think >" -%}
+      {%- endif -%}
+      {%- if think_close -%}
+        {%- set before_close = content.split(think_close)[0] -%}
+        {%- set reasoning = before_close.split(think_open)[-1] | trim -%}
+        {%- set content = content.split(think_close)[-1] | trim -%}
+      {%- endif -%}
+    {%- endif -%}
+    {%- set reasoning = reasoning | trim -%}
+
+    {{- "<|im_start|>assistant\n" -}}
+    {%- if reasoning and (preserve_thinking or loop.index0 > query_state.last_user_index) -%}
+      {{- "<think>\n" ~ reasoning ~ "\n</think>\n\n" -}}
+    {%- endif -%}
+    {{- content -}}
+
+    {%- if message.tool_calls is defined and message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping -%}
+      {%- for raw_call in message.tool_calls -%}
+        {%- set call = raw_call.function if raw_call.function is defined and raw_call.function is not none else raw_call -%}
+        {%- if tool_format == "json" -%}
+          {%- if content | trim or not loop.first -%}{{- "\n\n" -}}{%- endif -%}
+          {%- set serialized_args = "{}" -%}
+          {%- if call.arguments is defined and call.arguments is not none -%}
+            {%- if call.arguments is mapping -%}
+              {%- set serialized_args = call.arguments | tojson -%}
+            {%- elif call.arguments is string and call.arguments -%}
+              {%- set serialized_args = call.arguments -%}
+            {%- endif -%}
+          {%- endif -%}
+          {{- "<tool_call>\n{\"name\": " ~ (call.name | tojson) ~ ", \"arguments\": " ~ serialized_args ~ "}\n</tool_call>" -}}
+        {%- else -%}
+          {%- if content | trim or not loop.first -%}{{- "\n\n" -}}{%- endif -%}
+          {{- "<tool_call>\n<function=" ~ call.name ~ ">\n" -}}
+          {%- if call.arguments is defined and call.arguments is not none -%}
+            {%- if call.arguments is mapping -%}
+              {%- for arg_name in call.arguments -%}
+                {%- set arg_value = call.arguments[arg_name] -%}
+                {%- if arg_value is mapping or (arg_value is sequence and arg_value is not string) -%}
+                  {%- set arg_text = arg_value | tojson -%}
+                {%- else -%}
+                  {%- set arg_text = arg_value | string -%}
+                {%- endif -%}
+                {{- "<parameter=" ~ arg_name ~ ">\n" -}}
+                {%- if max_tool_arg_chars > 0 and arg_text | length > max_tool_arg_chars -%}
+                  {{- arg_text[:max_tool_arg_chars] ~ "\n[TRUNCATED]" -}}
+                {%- else -%}
+                  {{- arg_text -}}
+                {%- endif -%}
+                {{- "\n</parameter>\n" -}}
+              {%- endfor -%}
+            {%- elif call.arguments is string and call.arguments -%}
+              {{- call.arguments -}}
+            {%- endif -%}
+          {%- endif -%}
+          {{- "</function>\n</tool_call>" -}}
+        {%- endif -%}
+      {%- endfor -%}
+    {%- endif -%}
+
+    {# Some runtimes pass this kwarg to Jinja directly. New Transformers also trims via a sentinel. #}
+    {%- if not (loop.last and continue_final_message is defined and continue_final_message) -%}
+      {{- "<|im_end|>\n" -}}
+    {%- endif -%}
+
+  {%- elif role == "tool" -%}
+    {%- if state.previous_role != "tool" -%}
+      {{- "<|im_start|>user" -}}
+    {%- endif -%}
+    {%- if max_tool_response_chars > 0 and content | length > max_tool_response_chars -%}
+      {%- set content = content[:max_tool_response_chars] ~ "\n[TRUNCATED]" -%}
+    {%- endif -%}
+    {{- "\n<tool_response>\n" ~ content ~ "\n</tool_response>" -}}
+    {%- if loop.last or conversation[loop.index0 + 1].role != "tool" -%}
+      {{- "<|im_end|>\n" -}}
+    {%- endif -%}
+
+  {%- else -%}
+    {{- "<|im_start|>user\n[" ~ role ~ "]: " ~ content ~ "<|im_end|>\n" -}}
+  {%- endif -%}
+
+  {%- set state.previous_role = role -%}
+{%- endfor -%}
+
+{# Deterministic precedence: a prefilled final assistant turn wins, so the two flags
+   together can never emit an unterminated turn followed by a fresh assistant header. #}
+{%- set continuing_final = continue_final_message is defined and continue_final_message
+                           and conversation and conversation[-1].role == "assistant" -%}
+{%- if add_generation_prompt and not continuing_final -%}
+  {{- "<|im_start|>assistant\n" -}}
+  {%- if state.thinking -%}
+    {{- "<think>\n" -}}
+  {%- else -%}
+    {{- "<think>\n\n</think>\n\n" -}}
+  {%- endif -%}
+{%- endif -%}

config.json

@@ -0,0 +1,324 @@
+{
+  "architectures": [
+    "Qwen3_5ForConditionalGeneration"
+  ],
+  "dtype": "bfloat16",
+  "image_token_id": 248056,
+  "language_model_only": false,
+  "model_type": "qwen3_5",
+  "text_config": {
+    "attention_bias": false,
+    "attention_dropout": 0.0,
+    "attn_output_gate": true,
+    "bos_token_id": 248044,
+    "dtype": "bfloat16",
+    "eos_token_id": 248044,
+    "full_attention_interval": 4,
+    "head_dim": 256,
+    "hidden_act": "silu",
+    "hidden_size": 5120,
+    "initializer_range": 0.02,
+    "intermediate_size": 17408,
+    "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_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": 48,
+    "linear_value_head_dim": 128,
+    "mamba_ssm_dtype": "float32",
+    "max_position_embeddings": 262144,
+    "model_type": "qwen3_5_text",
+    "mtp_num_hidden_layers": 1,
+    "mtp_use_dedicated_embeddings": false,
+    "num_attention_heads": 24,
+    "num_hidden_layers": 64,
+    "num_key_value_heads": 4,
+    "output_gate_type": "swish",
+    "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": "5.9.0",
+  "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": 5120,
+    "patch_size": 16,
+    "spatial_merge_size": 2,
+    "temporal_patch_size": 2
+  },
+  "vision_end_token_id": 248054,
+  "vision_start_token_id": 248053,
+  "quantization_config": {
+    "config_groups": {
+      "group_0": {
+        "input_activations": {
+          "dynamic": false,
+          "num_bits": 4,
+          "type": "float",
+          "group_size": 16
+        },
+        "weights": {
+          "dynamic": false,
+          "num_bits": 4,
+          "type": "float",
+          "group_size": 16
+        },
+        "targets": [
+          "Linear"
+        ]
+      }
+    },
+    "ignore": [
+      "model.language_model.embed_tokens",
+      "model.language_model.layers.0.linear_attn.conv1d",
+      "model.language_model.layers.0.linear_attn.in_proj_a",
+      "model.language_model.layers.0.linear_attn.in_proj_b",
+      "model.language_model.layers.1.linear_attn.conv1d",
+      "model.language_model.layers.1.linear_attn.in_proj_a",
+      "model.language_model.layers.1.linear_attn.in_proj_b",
+      "model.language_model.layers.10.linear_attn.conv1d",
+      "model.language_model.layers.10.linear_attn.in_proj_a",
+      "model.language_model.layers.10.linear_attn.in_proj_b",
+      "model.language_model.layers.12.linear_attn.conv1d",
+      "model.language_model.layers.12.linear_attn.in_proj_a",
+      "model.language_model.layers.12.linear_attn.in_proj_b",
+      "model.language_model.layers.13.linear_attn.conv1d",
+      "model.language_model.layers.13.linear_attn.in_proj_a",
+      "model.language_model.layers.13.linear_attn.in_proj_b",
+      "model.language_model.layers.14.linear_attn.conv1d",
+      "model.language_model.layers.14.linear_attn.in_proj_a",
+      "model.language_model.layers.14.linear_attn.in_proj_b",
+      "model.language_model.layers.16.linear_attn.conv1d",
+      "model.language_model.layers.16.linear_attn.in_proj_a",
+      "model.language_model.layers.16.linear_attn.in_proj_b",
+      "model.language_model.layers.17.linear_attn.conv1d",
+      "model.language_model.layers.17.linear_attn.in_proj_a",
+      "model.language_model.layers.17.linear_attn.in_proj_b",
+      "model.language_model.layers.18.linear_attn.conv1d",
+      "model.language_model.layers.18.linear_attn.in_proj_a",
+      "model.language_model.layers.18.linear_attn.in_proj_b",
+      "model.language_model.layers.2.linear_attn.conv1d",
+      "model.language_model.layers.2.linear_attn.in_proj_a",
+      "model.language_model.layers.2.linear_attn.in_proj_b",
+      "model.language_model.layers.20.linear_attn.conv1d",
+      "model.language_model.layers.20.linear_attn.in_proj_a",
+      "model.language_model.layers.20.linear_attn.in_proj_b",
+      "model.language_model.layers.21.linear_attn.conv1d",
+      "model.language_model.layers.21.linear_attn.in_proj_a",
+      "model.language_model.layers.21.linear_attn.in_proj_b",
+      "model.language_model.layers.22.linear_attn.conv1d",
+      "model.language_model.layers.22.linear_attn.in_proj_a",
+      "model.language_model.layers.22.linear_attn.in_proj_b",
+      "model.language_model.layers.24.linear_attn.conv1d",
+      "model.language_model.layers.24.linear_attn.in_proj_a",
+      "model.language_model.layers.24.linear_attn.in_proj_b",
+      "model.language_model.layers.25.linear_attn.conv1d",
+      "model.language_model.layers.25.linear_attn.in_proj_a",
+      "model.language_model.layers.25.linear_attn.in_proj_b",
+      "model.language_model.layers.26.linear_attn.conv1d",
+      "model.language_model.layers.26.linear_attn.in_proj_a",
+      "model.language_model.layers.26.linear_attn.in_proj_b",
+      "model.language_model.layers.28.linear_attn.conv1d",
+      "model.language_model.layers.28.linear_attn.in_proj_a",
+      "model.language_model.layers.28.linear_attn.in_proj_b",
+      "model.language_model.layers.29.linear_attn.conv1d",
+      "model.language_model.layers.29.linear_attn.in_proj_a",
+      "model.language_model.layers.29.linear_attn.in_proj_b",
+      "model.language_model.layers.30.linear_attn.conv1d",
+      "model.language_model.layers.30.linear_attn.in_proj_a",
+      "model.language_model.layers.30.linear_attn.in_proj_b",
+      "model.language_model.layers.32.linear_attn.conv1d",
+      "model.language_model.layers.32.linear_attn.in_proj_a",
+      "model.language_model.layers.32.linear_attn.in_proj_b",
+      "model.language_model.layers.33.linear_attn.conv1d",
+      "model.language_model.layers.33.linear_attn.in_proj_a",
+      "model.language_model.layers.33.linear_attn.in_proj_b",
+      "model.language_model.layers.34.linear_attn.conv1d",
+      "model.language_model.layers.34.linear_attn.in_proj_a",
+      "model.language_model.layers.34.linear_attn.in_proj_b",
+      "model.language_model.layers.36.linear_attn.conv1d",
+      "model.language_model.layers.36.linear_attn.in_proj_a",
+      "model.language_model.layers.36.linear_attn.in_proj_b",
+      "model.language_model.layers.37.linear_attn.conv1d",
+      "model.language_model.layers.37.linear_attn.in_proj_a",
+      "model.language_model.layers.37.linear_attn.in_proj_b",
+      "model.language_model.layers.38.linear_attn.conv1d",
+      "model.language_model.layers.38.linear_attn.in_proj_a",
+      "model.language_model.layers.38.linear_attn.in_proj_b",
+      "model.language_model.layers.4.linear_attn.conv1d",
+      "model.language_model.layers.4.linear_attn.in_proj_a",
+      "model.language_model.layers.4.linear_attn.in_proj_b",
+      "model.language_model.layers.40.linear_attn.conv1d",
+      "model.language_model.layers.40.linear_attn.in_proj_a",
+      "model.language_model.layers.40.linear_attn.in_proj_b",
+      "model.language_model.layers.41.linear_attn.conv1d",
+      "model.language_model.layers.41.linear_attn.in_proj_a",
+      "model.language_model.layers.41.linear_attn.in_proj_b",
+      "model.language_model.layers.42.linear_attn.conv1d",
+      "model.language_model.layers.42.linear_attn.in_proj_a",
+      "model.language_model.layers.42.linear_attn.in_proj_b",
+      "model.language_model.layers.44.linear_attn.conv1d",
+      "model.language_model.layers.44.linear_attn.in_proj_a",
+      "model.language_model.layers.44.linear_attn.in_proj_b",
+      "model.language_model.layers.45.linear_attn.conv1d",
+      "model.language_model.layers.45.linear_attn.in_proj_a",
+      "model.language_model.layers.45.linear_attn.in_proj_b",
+      "model.language_model.layers.46.linear_attn.conv1d",
+      "model.language_model.layers.46.linear_attn.in_proj_a",
+      "model.language_model.layers.46.linear_attn.in_proj_b",
+      "model.language_model.layers.48.linear_attn.conv1d",
+      "model.language_model.layers.48.linear_attn.in_proj_a",
+      "model.language_model.layers.48.linear_attn.in_proj_b",
+      "model.language_model.layers.49.linear_attn.conv1d",
+      "model.language_model.layers.49.linear_attn.in_proj_a",
+      "model.language_model.layers.49.linear_attn.in_proj_b",
+      "model.language_model.layers.5.linear_attn.conv1d",
+      "model.language_model.layers.5.linear_attn.in_proj_a",
+      "model.language_model.layers.5.linear_attn.in_proj_b",
+      "model.language_model.layers.50.linear_attn.conv1d",
+      "model.language_model.layers.50.linear_attn.in_proj_a",
+      "model.language_model.layers.50.linear_attn.in_proj_b",
+      "model.language_model.layers.52.linear_attn.conv1d",
+      "model.language_model.layers.52.linear_attn.in_proj_a",
+      "model.language_model.layers.52.linear_attn.in_proj_b",
+      "model.language_model.layers.53.linear_attn.conv1d",
+      "model.language_model.layers.53.linear_attn.in_proj_a",
+      "model.language_model.layers.53.linear_attn.in_proj_b",
+      "model.language_model.layers.54.linear_attn.conv1d",
+      "model.language_model.layers.54.linear_attn.in_proj_a",
+      "model.language_model.layers.54.linear_attn.in_proj_b",
+      "model.language_model.layers.56.linear_attn.conv1d",
+      "model.language_model.layers.56.linear_attn.in_proj_a",
+      "model.language_model.layers.56.linear_attn.in_proj_b",
+      "model.language_model.layers.57.linear_attn.conv1d",
+      "model.language_model.layers.57.linear_attn.in_proj_a",
+      "model.language_model.layers.57.linear_attn.in_proj_b",
+      "model.language_model.layers.58.linear_attn.conv1d",
+      "model.language_model.layers.58.linear_attn.in_proj_a",
+      "model.language_model.layers.58.linear_attn.in_proj_b",
+      "model.language_model.layers.6.linear_attn.conv1d",
+      "model.language_model.layers.6.linear_attn.in_proj_a",
+      "model.language_model.layers.6.linear_attn.in_proj_b",
+      "model.language_model.layers.60.linear_attn.conv1d",
+      "model.language_model.layers.60.linear_attn.in_proj_a",
+      "model.language_model.layers.60.linear_attn.in_proj_b",
+      "model.language_model.layers.61.linear_attn.conv1d",
+      "model.language_model.layers.61.linear_attn.in_proj_a",
+      "model.language_model.layers.61.linear_attn.in_proj_b",
+      "model.language_model.layers.62.linear_attn.conv1d",
+      "model.language_model.layers.62.linear_attn.in_proj_a",
+      "model.language_model.layers.62.linear_attn.in_proj_b",
+      "model.language_model.layers.8.linear_attn.conv1d",
+      "model.language_model.layers.8.linear_attn.in_proj_a",
+      "model.language_model.layers.8.linear_attn.in_proj_b",
+      "model.language_model.layers.9.linear_attn.conv1d",
+      "model.language_model.layers.9.linear_attn.in_proj_a",
+      "model.language_model.layers.9.linear_attn.in_proj_b",
+      "model.visual*",
+      "mtp*",
+      "mtp.layers.0*"
+    ],
+    "quant_algo": "NVFP4",
+    "kv_cache_scheme": {
+      "dynamic": false,
+      "num_bits": 8,
+      "type": "float"
+    },
+    "producer": {
+      "name": "modelopt",
+      "version": "0.0.1.dev1+gc4129b6e0"
+    },
+    "quant_method": "modelopt"
+  }
+}

crc32.txt

@@ -0,0 +1,3 @@
+617cc98f  model-00001-of-00003.safetensors
+c159e96c  model-00002-of-00003.safetensors
+35bbe170  model-00003-of-00003.safetensors

generation_config.json

@@ -0,0 +1,13 @@
+{
+  "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,
+  "transformers_version": "5.9.0"
+}

hf_quant_config.json

@@ -0,0 +1,161 @@
+{
+  "producer": {
+    "name": "modelopt",
+    "version": "0.0.1.dev1+gc4129b6e0"
+  },
+  "quantization": {
+    "quant_algo": "NVFP4",
+    "kv_cache_quant_algo": "FP8",
+    "group_size": 16,
+    "exclude_modules": [
+      "model.language_model.embed_tokens",
+      "model.language_model.layers.0.linear_attn.conv1d",
+      "model.language_model.layers.0.linear_attn.in_proj_a",
+      "model.language_model.layers.0.linear_attn.in_proj_b",
+      "model.language_model.layers.1.linear_attn.conv1d",
+      "model.language_model.layers.1.linear_attn.in_proj_a",
+      "model.language_model.layers.1.linear_attn.in_proj_b",
+      "model.language_model.layers.10.linear_attn.conv1d",
+      "model.language_model.layers.10.linear_attn.in_proj_a",
+      "model.language_model.layers.10.linear_attn.in_proj_b",
+      "model.language_model.layers.12.linear_attn.conv1d",
+      "model.language_model.layers.12.linear_attn.in_proj_a",
+      "model.language_model.layers.12.linear_attn.in_proj_b",
+      "model.language_model.layers.13.linear_attn.conv1d",
+      "model.language_model.layers.13.linear_attn.in_proj_a",
+      "model.language_model.layers.13.linear_attn.in_proj_b",
+      "model.language_model.layers.14.linear_attn.conv1d",
+      "model.language_model.layers.14.linear_attn.in_proj_a",
+      "model.language_model.layers.14.linear_attn.in_proj_b",
+      "model.language_model.layers.16.linear_attn.conv1d",
+      "model.language_model.layers.16.linear_attn.in_proj_a",
+      "model.language_model.layers.16.linear_attn.in_proj_b",
+      "model.language_model.layers.17.linear_attn.conv1d",
+      "model.language_model.layers.17.linear_attn.in_proj_a",
+      "model.language_model.layers.17.linear_attn.in_proj_b",
+      "model.language_model.layers.18.linear_attn.conv1d",
+      "model.language_model.layers.18.linear_attn.in_proj_a",
+      "model.language_model.layers.18.linear_attn.in_proj_b",
+      "model.language_model.layers.2.linear_attn.conv1d",
+      "model.language_model.layers.2.linear_attn.in_proj_a",
+      "model.language_model.layers.2.linear_attn.in_proj_b",
+      "model.language_model.layers.20.linear_attn.conv1d",
+      "model.language_model.layers.20.linear_attn.in_proj_a",
+      "model.language_model.layers.20.linear_attn.in_proj_b",
+      "model.language_model.layers.21.linear_attn.conv1d",
+      "model.language_model.layers.21.linear_attn.in_proj_a",
+      "model.language_model.layers.21.linear_attn.in_proj_b",
+      "model.language_model.layers.22.linear_attn.conv1d",
+      "model.language_model.layers.22.linear_attn.in_proj_a",
+      "model.language_model.layers.22.linear_attn.in_proj_b",
+      "model.language_model.layers.24.linear_attn.conv1d",
+      "model.language_model.layers.24.linear_attn.in_proj_a",
+      "model.language_model.layers.24.linear_attn.in_proj_b",
+      "model.language_model.layers.25.linear_attn.conv1d",
+      "model.language_model.layers.25.linear_attn.in_proj_a",
+      "model.language_model.layers.25.linear_attn.in_proj_b",
+      "model.language_model.layers.26.linear_attn.conv1d",
+      "model.language_model.layers.26.linear_attn.in_proj_a",
+      "model.language_model.layers.26.linear_attn.in_proj_b",
+      "model.language_model.layers.28.linear_attn.conv1d",
+      "model.language_model.layers.28.linear_attn.in_proj_a",
+      "model.language_model.layers.28.linear_attn.in_proj_b",
+      "model.language_model.layers.29.linear_attn.conv1d",
+      "model.language_model.layers.29.linear_attn.in_proj_a",
+      "model.language_model.layers.29.linear_attn.in_proj_b",
+      "model.language_model.layers.30.linear_attn.conv1d",
+      "model.language_model.layers.30.linear_attn.in_proj_a",
+      "model.language_model.layers.30.linear_attn.in_proj_b",
+      "model.language_model.layers.32.linear_attn.conv1d",
+      "model.language_model.layers.32.linear_attn.in_proj_a",
+      "model.language_model.layers.32.linear_attn.in_proj_b",
+      "model.language_model.layers.33.linear_attn.conv1d",
+      "model.language_model.layers.33.linear_attn.in_proj_a",
+      "model.language_model.layers.33.linear_attn.in_proj_b",
+      "model.language_model.layers.34.linear_attn.conv1d",
+      "model.language_model.layers.34.linear_attn.in_proj_a",
+      "model.language_model.layers.34.linear_attn.in_proj_b",
+      "model.language_model.layers.36.linear_attn.conv1d",
+      "model.language_model.layers.36.linear_attn.in_proj_a",
+      "model.language_model.layers.36.linear_attn.in_proj_b",
+      "model.language_model.layers.37.linear_attn.conv1d",
+      "model.language_model.layers.37.linear_attn.in_proj_a",
+      "model.language_model.layers.37.linear_attn.in_proj_b",
+      "model.language_model.layers.38.linear_attn.conv1d",
+      "model.language_model.layers.38.linear_attn.in_proj_a",
+      "model.language_model.layers.38.linear_attn.in_proj_b",
+      "model.language_model.layers.4.linear_attn.conv1d",
+      "model.language_model.layers.4.linear_attn.in_proj_a",
+      "model.language_model.layers.4.linear_attn.in_proj_b",
+      "model.language_model.layers.40.linear_attn.conv1d",
+      "model.language_model.layers.40.linear_attn.in_proj_a",
+      "model.language_model.layers.40.linear_attn.in_proj_b",
+      "model.language_model.layers.41.linear_attn.conv1d",
+      "model.language_model.layers.41.linear_attn.in_proj_a",
+      "model.language_model.layers.41.linear_attn.in_proj_b",
+      "model.language_model.layers.42.linear_attn.conv1d",
+      "model.language_model.layers.42.linear_attn.in_proj_a",
+      "model.language_model.layers.42.linear_attn.in_proj_b",
+      "model.language_model.layers.44.linear_attn.conv1d",
+      "model.language_model.layers.44.linear_attn.in_proj_a",
+      "model.language_model.layers.44.linear_attn.in_proj_b",
+      "model.language_model.layers.45.linear_attn.conv1d",
+      "model.language_model.layers.45.linear_attn.in_proj_a",
+      "model.language_model.layers.45.linear_attn.in_proj_b",
+      "model.language_model.layers.46.linear_attn.conv1d",
+      "model.language_model.layers.46.linear_attn.in_proj_a",
+      "model.language_model.layers.46.linear_attn.in_proj_b",
+      "model.language_model.layers.48.linear_attn.conv1d",
+      "model.language_model.layers.48.linear_attn.in_proj_a",
+      "model.language_model.layers.48.linear_attn.in_proj_b",
+      "model.language_model.layers.49.linear_attn.conv1d",
+      "model.language_model.layers.49.linear_attn.in_proj_a",
+      "model.language_model.layers.49.linear_attn.in_proj_b",
+      "model.language_model.layers.5.linear_attn.conv1d",
+      "model.language_model.layers.5.linear_attn.in_proj_a",
+      "model.language_model.layers.5.linear_attn.in_proj_b",
+      "model.language_model.layers.50.linear_attn.conv1d",
+      "model.language_model.layers.50.linear_attn.in_proj_a",
+      "model.language_model.layers.50.linear_attn.in_proj_b",
+      "model.language_model.layers.52.linear_attn.conv1d",
+      "model.language_model.layers.52.linear_attn.in_proj_a",
+      "model.language_model.layers.52.linear_attn.in_proj_b",
+      "model.language_model.layers.53.linear_attn.conv1d",
+      "model.language_model.layers.53.linear_attn.in_proj_a",
+      "model.language_model.layers.53.linear_attn.in_proj_b",
+      "model.language_model.layers.54.linear_attn.conv1d",
+      "model.language_model.layers.54.linear_attn.in_proj_a",
+      "model.language_model.layers.54.linear_attn.in_proj_b",
+      "model.language_model.layers.56.linear_attn.conv1d",
+      "model.language_model.layers.56.linear_attn.in_proj_a",
+      "model.language_model.layers.56.linear_attn.in_proj_b",
+      "model.language_model.layers.57.linear_attn.conv1d",
+      "model.language_model.layers.57.linear_attn.in_proj_a",
+      "model.language_model.layers.57.linear_attn.in_proj_b",
+      "model.language_model.layers.58.linear_attn.conv1d",
+      "model.language_model.layers.58.linear_attn.in_proj_a",
+      "model.language_model.layers.58.linear_attn.in_proj_b",
+      "model.language_model.layers.6.linear_attn.conv1d",
+      "model.language_model.layers.6.linear_attn.in_proj_a",
+      "model.language_model.layers.6.linear_attn.in_proj_b",
+      "model.language_model.layers.60.linear_attn.conv1d",
+      "model.language_model.layers.60.linear_attn.in_proj_a",
+      "model.language_model.layers.60.linear_attn.in_proj_b",
+      "model.language_model.layers.61.linear_attn.conv1d",
+      "model.language_model.layers.61.linear_attn.in_proj_a",
+      "model.language_model.layers.61.linear_attn.in_proj_b",
+      "model.language_model.layers.62.linear_attn.conv1d",
+      "model.language_model.layers.62.linear_attn.in_proj_a",
+      "model.language_model.layers.62.linear_attn.in_proj_b",
+      "model.language_model.layers.8.linear_attn.conv1d",
+      "model.language_model.layers.8.linear_attn.in_proj_a",
+      "model.language_model.layers.8.linear_attn.in_proj_b",
+      "model.language_model.layers.9.linear_attn.conv1d",
+      "model.language_model.layers.9.linear_attn.in_proj_a",
+      "model.language_model.layers.9.linear_attn.in_proj_b",
+      "model.visual*",
+      "mtp*",
+      "mtp.layers.0*"
+    ]
+  }
+}

model-00001-of-00003.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cdd37b0e61eccc8a3d7d08f9d1a4f52856a9d88e4e8b42089bd18a970e3a01ec
+size 9972777720

model-00002-of-00003.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4b547449a2b23c6cd414da0cf65ff9d7e17ad9aa2b119beedcbba14f649eb1dd
+size 8048202912

model-00003-of-00003.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9ce944d534eabdd493076a3a52c7ebd31f41c135b340a1ea95c5a695e6f1f6b2
+size 744532384

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"
+}

processor_config.json

@@ -0,0 +1,60 @@
+{
+  "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,
+    "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"
+  }
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0997f410c57a1f4e53b09e4be8f4a172d90edd9564368fb0847030937229b9f3
+size 12809320

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|>",
+  "padding_side": "left",
+  "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+",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}

video_preprocessor_config.json

@@ -0,0 +1,21 @@
+{
+    "size": {
+        "longest_edge": 25165824,
+        "shortest_edge": 4096
+    },
+    "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",
+    "video_processor_type": "Qwen3VLVideoProcessor"
+}