Initial release: NVFP4 + MTP head bf16-restored, VLM kept (vision tower bf16), abliterated

.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 +1,205 @@
+---
+license: other
+base_model: huihui-ai/Huihui-Qwen3.6-27B-abliterated
+base_model_relation: quantized
+library_name: transformers
+tags:
+  - qwen3_5
+  - qwen3.6
+  - nvfp4
+  - quantized
+  - modelopt
+  - mtp
+  - speculative-decoding
+  - blackwell
+  - abliterated
+  - multimodal
+  - image-text-to-text
+  - vlm
+pipeline_tag: image-text-to-text
+language:
+  - en
+  - zh
+  - ja
+  - ko
+  - fr
+  - de
+  - es
+  - it
+  - pt
+  - ru
+  - ar
+---
+
+# Huihui-Qwen3.6-27B-abliterated-NVFP4-MTP
+
+NVFP4-quantized **multimodal** abliterated sibling of [`Qwen/Qwen3.6-27B`](https://huggingface.co/Qwen/Qwen3.6-27B), with the **MTP (Multi-Token Prediction) head restored in bf16** so speculative decoding works.
+
+Vision tower is **kept** (in bf16 — image and video input still work).
+
+## Sibling repos
+
+| | This repo | Text-only sibling | Original VLM (compressed-tensors) |
+|---|---|---|---|
+| Repo | `Huihui-Qwen3.6-27B-abliterated-NVFP4-MTP` | [`Qwen3.6-27B-Text-NVFP4-MTP`](https://huggingface.co/sakamakismile/Qwen3.6-27B-Text-NVFP4-MTP) | [`Huihui-Qwen3.6-27B-abliterated-NVFP4`](https://huggingface.co/sakamakismile/Huihui-Qwen3.6-27B-abliterated-NVFP4) |
+| Vision input | **✅ image + video** | ❌ text-only | ✅ |
+| Quantization format | **`modelopt`** (vLLM SM120 native path) | `modelopt` | `compressed-tensors` |
+| MTP head | **✅ bf16, working** | ✅ bf16, working | ❌ dropped → 0% acceptance |
+| Abliterated | **✅ (huihui-ai base)** | ❌ | ✅ |
+
+## What's different from the original `Huihui-Qwen3.6-27B-abliterated-NVFP4`
+
+The original repo was quantized with `llmcompressor` to `compressed-tensors` format
+and had the same problem we hit on the official Qwen repo: `AutoModelForCausalLM`
+silently drops the MTP head during export, leading to 0% draft acceptance, **and**
+`compressed-tensors` NVFP4 takes a slower fallback path on Blackwell SM120 in
+vLLM 0.19.x.
+
+This repo:
+
+1. **Quantization format**: `modelopt` (native NVFP4 GEMM via
+   `FlashInferCutlassNvFp4LinearKernel` on Blackwell)
+2. **MTP head**: 15 `mtp.*` tensors (~850 MB, bf16) re-grafted from the original
+   bf16 base, and added to `quantization_config.ignore`
+3. **Vision tower**: **kept** in bf16 (`*visual*` on the ignore list — image
+   encoder integrity preserved)
+4. **Mamba/SSM convs**: `*linear_attn.conv1d*` is on `modelopt`'s default ignore
+   list, so the hybrid layer convolutions stay in bf16
+5. **Abliteration**: inherited from the upstream [`huihui-ai/Huihui-Qwen3.6-27B-abliterated`](https://huggingface.co/huihui-ai/Huihui-Qwen3.6-27B-abliterated) base — no
+   additional alignment pass
+
+## Why "Unsensor"?
+
+This is the abliterated counterpart of our text-only release. The intent (per the
+maintainer's [philosophy](https://huggingface.co/sakamakismile)) is **not "remove
+the chains" but "remove the colored glasses"** — let the model observe and reason
+neutrally, without the strong refusal-shaped priors learned during alignment.
+You're expected to use it responsibly.
+
+## Quantization details
+
+- **Base**: `huihui-ai/Huihui-Qwen3.6-27B-abliterated` (bf16, 27.78B params,
+  hybrid linear-attn + full-attn, 64 layers, 1 MTP layer)
+- **Quantizer**: `nvidia-modelopt` 0.43.0 with `NVFP4_DEFAULT_CFG`
+- **Calibration**: 20 samples from `neuralmagic/calibration` (LLM split),
+  max_seq_len 8192
+- **Ignored from quantization** (kept in bf16):
+  - `lm_head`
+  - All `model.visual.*` (vision tower bf16-preserved)
+  - All `*linear_attn.conv1d*` (Mamba-style SSM convolutions, 48 of 64 layers)
+  - All `mtp.*` modules (15 tensors, ~850 MB bf16)
+  - Other `NVFP4_DEFAULT_CFG` defaults (router, mlp.gate, output_layer …)
+
+## Usage with vLLM (Blackwell, SM120)
+
+### With image input + MTP speculative decoding
+
+```bash
+vllm serve sakamakismile/Huihui-Qwen3.6-27B-abliterated-NVFP4-MTP \
+    --trust-remote-code \
+    --gpu-memory-utilization 0.85 \
+    --max-model-len 8192 \
+    --quantization modelopt \
+    --speculative-config '{"method":"qwen3_5_mtp","num_speculative_tokens":1}'
+```
+
+Then send an image-aware request:
+
+```bash
+curl http://localhost:8000/v1/chat/completions -H 'Content-Type: application/json' -d '{
+  "model": "sakamakismile/Huihui-Qwen3.6-27B-abliterated-NVFP4-MTP",
+  "messages": [{
+    "role": "user",
+    "content": [
+      {"type": "image_url", "image_url": {"url": "https://example.com/photo.jpg"}},
+      {"type": "text", "text": "Describe what you see in this image."}
+    ]
+  }],
+  "max_tokens": 400
+}'
+```
+
+`num_speculative_tokens: 1` because the model has a single MTP layer
+(`mtp_num_hidden_layers=1`). The handler `qwen3_5_mtp` is internally
+normalized to `mtp` by current vLLM (deprecated-name warning is harmless).
+
+### Multi-instance + KV FP8 (high-throughput serving)
+
+For aggregate-throughput serving on a 6-GPU Blackwell box, two instances per
+GPU with `--kv-cache-dtype fp8` is the recommended layout:
+
+```bash
+CUDA_VISIBLE_DEVICES=0 vllm serve <repo> \
+    --gpu-memory-utilization 0.45 \
+    --kv-cache-dtype fp8 \
+    --max-model-len 8192 \
+    --quantization modelopt \
+    --speculative-config '{"method":"qwen3_5_mtp","num_speculative_tokens":1}' \
+    --port 8002 &
+# repeat for ports 8003-8013 across CUDA_VISIBLE_DEVICES 0..5
+```
+
+KV FP8 is verified to introduce no measurable quality regression on the
+Qwen3.5/3.6 family; halving the KV footprint approximately doubles the
+in-flight slot budget per instance.
+
+## Verified locally (RTX PRO 6000 Blackwell, vLLM 0.19.1rc1)
+
+Single request, T=0.7, 2000-token long-form decode across 4 domains:
+
+| Domain | tok/s |
+|---|---|
+| Technical (CS) | 87.0 |
+| Literary (Japanese) | 76.4 |
+| Reasoning (math) | 90.6 |
+| Code review | 90.5 |
+| **mean** | **86.9** |
+
+- **MTP acceptance**: 81.1% on long-form (T=0.7), 86.8% on short prompts (T=0)
+- **Image input** (Pexels cat / HF logo / code screenshot / public domain art):
+  77–84 tok/s with image preprocess included; output is coherent and
+  domain-aware
+
+### Aggregate throughput (6 GPUs, 1 instance per GPU, max-num-seqs=2)
+
+12 concurrent requests (× 500 tokens, T=0.7) round-robined across 6 endpoints:
+
+| metric | value |
+|---|---|
+| concurrency | 12 (= 6 GPUs × 2 in-flight) |
+| successful requests | 12 / 12 |
+| wall time | 9.0 s |
+| **aggregate tok/s** | **648.4** |
+| avg per-req tok/s | 57.3 |
+| MTP acceptance | 80.1% |
+
+That's ~7.5× the single-request decode rate, which is the realistic ceiling
+for a 6 × RTX PRO 6000 Blackwell box serving this model concurrently —
+each user lands on a GPU that handles up to 2 in-flight requests at a
+time via vLLM's continuous batching. (Co-resident multi-instance per GPU
+was attempted and rejected: vLLM V1 cannot share VRAM accounting across
+two processes on the same GPU, so a 12-instance / 24-in-flight layout
+hits CUDA OOM during cuda-graph capture; this is upstream-known. RTX PRO
+6000 Blackwell *Workstation* does not expose MIG either.)
+
+## Hardware target
+
+Built and tested on **NVIDIA RTX PRO 6000 Blackwell (SM120)**. Should also work
+on **RTX 5090** and other Blackwell consumer/workstation cards with sufficient
+VRAM (the model is roughly 15 GB NVFP4 weights + ~5 GB bf16 vision/MTP/SSM/
+lm_head ≈ 20.6 GB on disk, ≈ 21 GB at load with KV cache room on top).
+
+## Acknowledgements
+
+- [`huihui-ai`](https://huggingface.co/huihui-ai) — for the abliterated base
+- [`Qwen`](https://huggingface.co/Qwen) — for the original Qwen3.6-27B
+- [`osoleve`](https://huggingface.co/osoleve) — for the MTP-restoration recipe
+  on Qwen3.5
+- [`nvidia-modelopt`](https://github.com/NVIDIA/TensorRT-Model-Optimizer) team
+- The reporters of Discussions #5 and #7 on the original repo — for catching
+  the issues cleanly
+
+## License
+
+Inherited from the upstream Huihui base (other / abliterated terms apply —
+see [huihui-ai/Huihui-Qwen3.6-27B-abliterated](https://huggingface.co/huihui-ai/Huihui-Qwen3.6-27B-abliterated)).

chat_template.jinja

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

config.json

@@ -0,0 +1,236 @@
+{
+  "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": "5.5.4",
+  "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",
+    "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": [
+      "lm_head",
+      "model.language_model.layers.0.linear_attn.conv1d",
+      "model.language_model.layers.1.linear_attn.conv1d",
+      "model.language_model.layers.10.linear_attn.conv1d",
+      "model.language_model.layers.12.linear_attn.conv1d",
+      "model.language_model.layers.13.linear_attn.conv1d",
+      "model.language_model.layers.14.linear_attn.conv1d",
+      "model.language_model.layers.16.linear_attn.conv1d",
+      "model.language_model.layers.17.linear_attn.conv1d",
+      "model.language_model.layers.18.linear_attn.conv1d",
+      "model.language_model.layers.2.linear_attn.conv1d",
+      "model.language_model.layers.20.linear_attn.conv1d",
+      "model.language_model.layers.21.linear_attn.conv1d",
+      "model.language_model.layers.22.linear_attn.conv1d",
+      "model.language_model.layers.24.linear_attn.conv1d",
+      "model.language_model.layers.25.linear_attn.conv1d",
+      "model.language_model.layers.26.linear_attn.conv1d",
+      "model.language_model.layers.28.linear_attn.conv1d",
+      "model.language_model.layers.29.linear_attn.conv1d",
+      "model.language_model.layers.30.linear_attn.conv1d",
+      "model.language_model.layers.32.linear_attn.conv1d",
+      "model.language_model.layers.33.linear_attn.conv1d",
+      "model.language_model.layers.34.linear_attn.conv1d",
+      "model.language_model.layers.36.linear_attn.conv1d",
+      "model.language_model.layers.37.linear_attn.conv1d",
+      "model.language_model.layers.38.linear_attn.conv1d",
+      "model.language_model.layers.4.linear_attn.conv1d",
+      "model.language_model.layers.40.linear_attn.conv1d",
+      "model.language_model.layers.41.linear_attn.conv1d",
+      "model.language_model.layers.42.linear_attn.conv1d",
+      "model.language_model.layers.44.linear_attn.conv1d",
+      "model.language_model.layers.45.linear_attn.conv1d",
+      "model.language_model.layers.46.linear_attn.conv1d",
+      "model.language_model.layers.48.linear_attn.conv1d",
+      "model.language_model.layers.49.linear_attn.conv1d",
+      "model.language_model.layers.5.linear_attn.conv1d",
+      "model.language_model.layers.50.linear_attn.conv1d",
+      "model.language_model.layers.52.linear_attn.conv1d",
+      "model.language_model.layers.53.linear_attn.conv1d",
+      "model.language_model.layers.54.linear_attn.conv1d",
+      "model.language_model.layers.56.linear_attn.conv1d",
+      "model.language_model.layers.57.linear_attn.conv1d",
+      "model.language_model.layers.58.linear_attn.conv1d",
+      "model.language_model.layers.6.linear_attn.conv1d",
+      "model.language_model.layers.60.linear_attn.conv1d",
+      "model.language_model.layers.61.linear_attn.conv1d",
+      "model.language_model.layers.62.linear_attn.conv1d",
+      "model.language_model.layers.8.linear_attn.conv1d",
+      "model.language_model.layers.9.linear_attn.conv1d",
+      "model.visual*",
+      "mtp.fc",
+      "mtp.layers.0.input_layernorm",
+      "mtp.layers.0.mlp.down_proj",
+      "mtp.layers.0.mlp.gate_proj",
+      "mtp.layers.0.mlp.up_proj",
+      "mtp.layers.0.post_attention_layernorm",
+      "mtp.layers.0.self_attn.k_norm",
+      "mtp.layers.0.self_attn.k_proj",
+      "mtp.layers.0.self_attn.o_proj",
+      "mtp.layers.0.self_attn.q_norm",
+      "mtp.layers.0.self_attn.q_proj",
+      "mtp.layers.0.self_attn.v_proj",
+      "mtp.norm",
+      "mtp.pre_fc_norm_embedding",
+      "mtp.pre_fc_norm_hidden"
+    ],
+    "quant_algo": "NVFP4",
+    "producer": {
+      "name": "modelopt",
+      "version": "0.43.0"
+    },
+    "quant_method": "modelopt"
+  }
+}

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": "5.5.4"
+}

hf_quant_config.json

@@ -0,0 +1,63 @@
+{
+    "producer": {
+        "name": "modelopt",
+        "version": "0.43.0"
+    },
+    "quantization": {
+        "quant_algo": "NVFP4",
+        "kv_cache_quant_algo": null,
+        "group_size": 16,
+        "exclude_modules": [
+            "lm_head",
+            "model.language_model.layers.0.linear_attn.conv1d",
+            "model.language_model.layers.1.linear_attn.conv1d",
+            "model.language_model.layers.10.linear_attn.conv1d",
+            "model.language_model.layers.12.linear_attn.conv1d",
+            "model.language_model.layers.13.linear_attn.conv1d",
+            "model.language_model.layers.14.linear_attn.conv1d",
+            "model.language_model.layers.16.linear_attn.conv1d",
+            "model.language_model.layers.17.linear_attn.conv1d",
+            "model.language_model.layers.18.linear_attn.conv1d",
+            "model.language_model.layers.2.linear_attn.conv1d",
+            "model.language_model.layers.20.linear_attn.conv1d",
+            "model.language_model.layers.21.linear_attn.conv1d",
+            "model.language_model.layers.22.linear_attn.conv1d",
+            "model.language_model.layers.24.linear_attn.conv1d",
+            "model.language_model.layers.25.linear_attn.conv1d",
+            "model.language_model.layers.26.linear_attn.conv1d",
+            "model.language_model.layers.28.linear_attn.conv1d",
+            "model.language_model.layers.29.linear_attn.conv1d",
+            "model.language_model.layers.30.linear_attn.conv1d",
+            "model.language_model.layers.32.linear_attn.conv1d",
+            "model.language_model.layers.33.linear_attn.conv1d",
+            "model.language_model.layers.34.linear_attn.conv1d",
+            "model.language_model.layers.36.linear_attn.conv1d",
+            "model.language_model.layers.37.linear_attn.conv1d",
+            "model.language_model.layers.38.linear_attn.conv1d",
+            "model.language_model.layers.4.linear_attn.conv1d",
+            "model.language_model.layers.40.linear_attn.conv1d",
+            "model.language_model.layers.41.linear_attn.conv1d",
+            "model.language_model.layers.42.linear_attn.conv1d",
+            "model.language_model.layers.44.linear_attn.conv1d",
+            "model.language_model.layers.45.linear_attn.conv1d",
+            "model.language_model.layers.46.linear_attn.conv1d",
+            "model.language_model.layers.48.linear_attn.conv1d",
+            "model.language_model.layers.49.linear_attn.conv1d",
+            "model.language_model.layers.5.linear_attn.conv1d",
+            "model.language_model.layers.50.linear_attn.conv1d",
+            "model.language_model.layers.52.linear_attn.conv1d",
+            "model.language_model.layers.53.linear_attn.conv1d",
+            "model.language_model.layers.54.linear_attn.conv1d",
+            "model.language_model.layers.56.linear_attn.conv1d",
+            "model.language_model.layers.57.linear_attn.conv1d",
+            "model.language_model.layers.58.linear_attn.conv1d",
+            "model.language_model.layers.6.linear_attn.conv1d",
+            "model.language_model.layers.60.linear_attn.conv1d",
+            "model.language_model.layers.61.linear_attn.conv1d",
+            "model.language_model.layers.62.linear_attn.conv1d",
+            "model.language_model.layers.8.linear_attn.conv1d",
+            "model.language_model.layers.9.linear_attn.conv1d",
+            "model.visual*"
+        ]
+    }
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7b213dc5d299dbb62da1a58ff0e90370cae392de0b4ae48d7ba023323c8992b5
+size 20559273880

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 sha256:128d23d7eac9325f73bc0e8c145dfb0ecac6230d455e80ab95423184e262f101
+size 19989442

tokenizer_config.json

@@ -0,0 +1,31 @@
+{
+  "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,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "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"
+}