Upload InternVL3.5-4B FP8 Dynamic checkpoint and vLLM guide

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

NOTICE

@@ -0,0 +1,9 @@
+InternVL3.5-4B-HF FP8 Dynamic
+
+This model is a quantized derivative of:
+OpenGVLab/InternVL3_5-4B-HF
+https://huggingface.co/OpenGVLab/InternVL3_5-4B-HF
+
+The upstream project is licensed under the Apache License 2.0.
+The quantized checkpoint preserves the upstream model architecture and files,
+with selected language decoder Linear weights compressed using FP8_DYNAMIC.

QUANTIZATION_INFO.json

@@ -0,0 +1,22 @@
+{
+  "source_directory": "/home/bellock/projects/internvl35-fp8/models/InternVL3_5-4B-HF",
+  "output_directory": "/home/bellock/projects/internvl35-fp8/models/InternVL3_5-4B-FP8-Dynamic",
+  "source_revision": "model_id=OpenGVLab/InternVL3_5-4B-HF\nrevision=6bd4487402110ef9889ba50eb7aefeb302526fed",
+  "quantization_scheme": "FP8_DYNAMIC",
+  "target_module_type": "Linear",
+  "target_linear_count": 252,
+  "protected_linear_count": 147,
+  "ignored_patterns": [
+    ".*vision_tower.*",
+    ".*multi_modal_projector.*",
+    ".*lm_head.*"
+  ],
+  "source_dtype": "torch.bfloat16",
+  "calibration_dataset": null,
+  "load_seconds": 0.8194205139998303,
+  "quantization_and_save_seconds": 17.089425016999485,
+  "python_version": "3.12.3",
+  "torch_version": "2.12.0+cu132",
+  "transformers_version": "5.10.1",
+  "llmcompressor_version": "0.12.0.1"
+}

README.md

@@ -0,0 +1,270 @@
+---
+license: apache-2.0
+library_name: transformers
+pipeline_tag: image-text-to-text
+base_model: OpenGVLab/InternVL3_5-4B-HF
+base_model_relation: quantized
+language:
+  - en
+  - zh
+  - ko
+tags:
+  - internvl
+  - internvl3.5
+  - vision-language
+  - multimodal
+  - vllm
+  - compressed-tensors
+  - fp8
+  - w8a16
+  - ampere
+  - wsl2
+---
+
+# InternVL3.5-4B-HF FP8 Dynamic
+
+This repository contains a compressed-tensors FP8 Dynamic quantization of
+[OpenGVLab/InternVL3_5-4B-HF](https://huggingface.co/OpenGVLab/InternVL3_5-4B-HF),
+prepared for memory-conscious vLLM serving.
+
+## Important runtime note
+
+The checkpoint stores the language decoder Linear weights in FP8 E4M3 format.
+On an NVIDIA Ampere GPU such as the RTX 3070, vLLM 0.26.0 serves these weights
+through its W8A16 FP8 path (Humming kernel): weights are compressed to 8-bit,
+while activations run in FP16. This is primarily a VRAM-saving configuration;
+a speedup is not guaranteed on Ampere.
+
+## Quantization scope
+
+Quantized:
+
+- 252 language decoder `Linear` modules
+- Scheme: `FP8_DYNAMIC`
+- Weight format: FP8 E4M3
+- Activation scaling: dynamic per token at runtime
+- Calibration dataset: not required
+
+Kept in BF16:
+
+- Vision tower
+- Multimodal projector
+- Input embeddings
+- `lm_head`
+- Normalization layers and other protected parameters
+
+The checkpoint was generated from the base-model revision:
+
+```text
+6bd4487402110ef9889ba50eb7aefeb302526fed
+```
+
+See [`quantization/recipe.py`](./quantization/recipe.py) for the compression recipe.
+
+## Verified environment
+
+The following setup was used for the initial serving validation:
+
+| Component | Version / value |
+|---|---|
+| GPU | NVIDIA GeForce RTX 3070 8GB |
+| Host | Windows + WSL2 |
+| WSL distribution | Ubuntu 24.04 |
+| NVIDIA driver | 591.86 |
+| vLLM | 0.26.0 |
+| PyTorch | 2.11.0+cu130 |
+| Transformers | 5.14.1 |
+| Quantization backend | compressed-tensors |
+| vLLM runner | V1 |
+| Attention | FlashAttention 2 |
+| Max context used in validation | 2,048 tokens |
+| Maximum images per request | 4 |
+
+Observed during startup with the verified preset:
+
+- Model weight memory: approximately **5.51 GiB**
+- Available KV-cache memory: approximately **0.59 GiB**
+- GPU KV-cache capacity: **4,288 tokens**
+- vLLM target memory at `--gpu-memory-utilization 0.82`: approximately **6.56 GiB**
+
+Windows graphics applications consume additional VRAM outside the vLLM process.
+The total value shown by Windows `nvidia-smi` can therefore be higher.
+
+## Quick start: WSL2 + RTX 3070
+
+### 1. Install system build requirements
+
+Humming compiles a small runtime extension on first use.
+
+```bash
+sudo apt update
+sudo apt install -y build-essential python3.12-dev
+```
+
+Do not install a Linux NVIDIA display driver inside WSL2. The Windows NVIDIA
+driver exposes `libcuda.so` under `/usr/lib/wsl/lib`.
+
+### 2. Create the Python environment
+
+Install [`uv`](https://docs.astral.sh/uv/) first when it is not already available.
+
+```bash
+uv venv --python 3.12 .venv-vllm
+source .venv-vllm/bin/activate
+uv pip install "vllm==0.26.0" hf_xet
+```
+
+### 3. Validate the environment
+
+```bash
+./scripts/check_wsl_runtime.sh
+```
+
+### 4. Start the server directly from Hugging Face
+
+```bash
+source .venv-vllm/bin/activate
+./scripts/start_vllm_wsl_rtx3070.sh
+```
+
+The default model ID is:
+
+```text
+hsmin92/internvl35-fp8
+```
+
+Successful startup ends with:
+
+```text
+Application startup complete.
+```
+
+The OpenAI-compatible endpoint is then available at:
+
+```text
+http://127.0.0.1:8000/v1
+```
+
+### Optional: lower KV-cache preset
+
+The default script uses the startup configuration that was validated first.
+To reduce the fixed KV-cache allocation, set both values together:
+
+```bash
+GPU_MEMORY_UTILIZATION=0.80 \
+KV_CACHE_MEMORY_BYTES=384M \
+./scripts/start_vllm_wsl_rtx3070.sh
+```
+
+The fixed KV-cache option does not replace `GPU_MEMORY_UTILIZATION`; both are
+needed by vLLM 0.26.0. Validate this preset on the target machine because the
+Windows desktop and browser processes also consume VRAM.
+
+## Runtime options
+
+The startup script accepts environment variables:
+
+| Variable | Default | Description |
+|---|---:|---|
+| `MODEL_ID` | `hsmin92/internvl35-fp8` | Hub model ID or local model path |
+| `SERVED_MODEL_NAME` | `internvl35-fp8` | Name exposed by the API |
+| `HOST` | `127.0.0.1` | Listen address |
+| `PORT` | `8000` | Listen port |
+| `MAX_MODEL_LEN` | `2048` | Total context budget, including image and output tokens |
+| `MAX_NUM_SEQS` | `1` | Maximum concurrent sequences |
+| `MAX_IMAGES` | `4` | Maximum images in one request |
+| `GPU_MEMORY_UTILIZATION` | `0.82` | vLLM GPU-memory target |
+| `KV_CACHE_MEMORY_BYTES` | unset | Optional fixed KV-cache size such as `384M` |
+
+Example:
+
+```bash
+PORT=8100 MAX_IMAGES=1 MAX_MODEL_LEN=1024 \
+./scripts/start_vllm_wsl_rtx3070.sh
+```
+
+## API tests
+
+### Health and model list
+
+```bash
+curl -s http://127.0.0.1:8000/health
+curl -s http://127.0.0.1:8000/v1/models | python3 -m json.tool
+```
+
+### Text request
+
+```bash
+./examples/chat_text.sh
+```
+
+### Local image request
+
+```bash
+python examples/chat_image.py /path/to/image.jpg \
+  "Describe the scene and list any safety-relevant events."
+```
+
+The image client sends the local image as a base64 data URL and uses only the
+Python standard library.
+
+## Native Linux and other GPUs
+
+The WSL2 script deliberately applies compatibility settings required by the
+validated RTX 3070 environment:
+
+- `VLLM_USE_V2_MODEL_RUNNER=0` because the V2 runner required UVA in this WSL setup.
+- `VLLM_USE_FLASHINFER_SAMPLER=0` because FlashInfer sampling JIT required `nvcc`.
+- `/usr/lib/wsl/lib` is added to the compile and runtime linker paths.
+- pip-installed CUDA NVRTC libraries are added to `LD_LIBRARY_PATH`.
+- `--enforce-eager` disables CUDA graphs and `torch.compile` for compatibility.
+
+Native Linux systems with a full CUDA Toolkit or newer GPUs may not need these
+workarounds. Start from the documented script, then remove compatibility flags
+one at a time and validate output quality, memory, and stability.
+
+## Intended use
+
+This model is suitable for experimentation with:
+
+- Image understanding
+- Multi-image comparison
+- CCTV frame summarization
+- Visual question answering
+- OpenAI-compatible multimodal API integration
+
+For video analysis on an 8GB GPU, sample a small number of frames externally,
+resize them appropriately, and send the frames as multiple images rather than
+passing every frame of a video.
+
+## Limitations
+
+- This is a quantized derivative, not an independently trained model.
+- The vision tower and output head remain BF16 and account for a meaningful
+  portion of the loaded weights.
+- FP8 on Ampere is served through a W8A16 compatibility kernel rather than
+  native FP8 Tensor Core execution.
+- A comprehensive quality benchmark against the BF16 base model has not yet
+  been published in this repository.
+- The first server start may compile and cache Humming runtime components.
+- VRAM figures depend on driver, desktop applications, context length,
+  multimodal limits, and vLLM version.
+
+## Attribution and license
+
+This repository is a quantized derivative of
+[OpenGVLab/InternVL3_5-4B-HF](https://huggingface.co/OpenGVLab/InternVL3_5-4B-HF).
+The original project and this derivative are distributed under the Apache-2.0
+license. Review the upstream model card for the original training details,
+limitations, and citation information.
+
+## Citation
+
+```bibtex
+@article{wang2025internvl3_5,
+  title={InternVL3.5: Advancing Open-Source Multimodal Models in Versatility, Reasoning, and Efficiency},
+  author={Wang, Weiyun and Gao, Zhangwei and Gu, Lixin and Pu, Hengjun and Cui, Long and Wei, Xingguang and Liu, Zhaoyang and Jing, Linglin and Ye, Shenglong and Shao, Jie and others},
+  journal={arXiv preprint arXiv:2508.18265},
+  year={2025}
+}
+```

chat_template.jinja

@@ -0,0 +1,6 @@
+{% for message in messages %}{{'<|im_start|>' + message['role'] + '
+'}}{% if message['content'] is string %}{{ message['content'] }}{% else %}{% for content in message['content'] %}{% if content['type'] == 'image' %}{{ '<IMG_CONTEXT>
+' }}{% elif content['type'] == 'video' %}{{ '<video>
+' }}{% elif content['type'] == 'text' %}{{ content['text'] }}{% endif %}{% endfor %}{% endif %}{{'<|im_end|>
+'}}{% endfor %}{% if add_generation_prompt %}{{'<|im_start|>assistant
+' }}{% endif %}

config.json

@@ -0,0 +1,322 @@
+{
+  "architectures": [
+    "InternVLForConditionalGeneration"
+  ],
+  "downsample_ratio": 0.5,
+  "dtype": "bfloat16",
+  "image_seq_length": 256,
+  "image_token_id": 151671,
+  "model_type": "internvl",
+  "projector_hidden_act": "gelu",
+  "quantization_config": {
+    "config_groups": {
+      "group_0": {
+        "format": "float-quantized",
+        "input_activations": {
+          "actorder": null,
+          "block_structure": null,
+          "dynamic": true,
+          "group_size": null,
+          "num_bits": 8,
+          "observer": null,
+          "observer_kwargs": {},
+          "scale_dtype": null,
+          "strategy": "token",
+          "symmetric": true,
+          "type": "float",
+          "zp_dtype": null
+        },
+        "output_activations": null,
+        "targets": [
+          "Linear"
+        ],
+        "weights": {
+          "actorder": null,
+          "block_structure": null,
+          "dynamic": false,
+          "group_size": null,
+          "num_bits": 8,
+          "observer": "memoryless_minmax",
+          "observer_kwargs": {},
+          "scale_dtype": null,
+          "strategy": "channel",
+          "symmetric": true,
+          "type": "float",
+          "zp_dtype": null
+        }
+      }
+    },
+    "format": "float-quantized",
+    "global_compression_ratio": null,
+    "ignore": [
+      "model.vision_tower.encoder.layer.0.attention.q_proj",
+      "model.vision_tower.encoder.layer.0.attention.k_proj",
+      "model.vision_tower.encoder.layer.0.attention.v_proj",
+      "model.vision_tower.encoder.layer.0.attention.projection_layer",
+      "model.vision_tower.encoder.layer.0.mlp.fc1",
+      "model.vision_tower.encoder.layer.0.mlp.fc2",
+      "model.vision_tower.encoder.layer.1.attention.q_proj",
+      "model.vision_tower.encoder.layer.1.attention.k_proj",
+      "model.vision_tower.encoder.layer.1.attention.v_proj",
+      "model.vision_tower.encoder.layer.1.attention.projection_layer",
+      "model.vision_tower.encoder.layer.1.mlp.fc1",
+      "model.vision_tower.encoder.layer.1.mlp.fc2",
+      "model.vision_tower.encoder.layer.2.attention.q_proj",
+      "model.vision_tower.encoder.layer.2.attention.k_proj",
+      "model.vision_tower.encoder.layer.2.attention.v_proj",
+      "model.vision_tower.encoder.layer.2.attention.projection_layer",
+      "model.vision_tower.encoder.layer.2.mlp.fc1",
+      "model.vision_tower.encoder.layer.2.mlp.fc2",
+      "model.vision_tower.encoder.layer.3.attention.q_proj",
+      "model.vision_tower.encoder.layer.3.attention.k_proj",
+      "model.vision_tower.encoder.layer.3.attention.v_proj",
+      "model.vision_tower.encoder.layer.3.attention.projection_layer",
+      "model.vision_tower.encoder.layer.3.mlp.fc1",
+      "model.vision_tower.encoder.layer.3.mlp.fc2",
+      "model.vision_tower.encoder.layer.4.attention.q_proj",
+      "model.vision_tower.encoder.layer.4.attention.k_proj",
+      "model.vision_tower.encoder.layer.4.attention.v_proj",
+      "model.vision_tower.encoder.layer.4.attention.projection_layer",
+      "model.vision_tower.encoder.layer.4.mlp.fc1",
+      "model.vision_tower.encoder.layer.4.mlp.fc2",
+      "model.vision_tower.encoder.layer.5.attention.q_proj",
+      "model.vision_tower.encoder.layer.5.attention.k_proj",
+      "model.vision_tower.encoder.layer.5.attention.v_proj",
+      "model.vision_tower.encoder.layer.5.attention.projection_layer",
+      "model.vision_tower.encoder.layer.5.mlp.fc1",
+      "model.vision_tower.encoder.layer.5.mlp.fc2",
+      "model.vision_tower.encoder.layer.6.attention.q_proj",
+      "model.vision_tower.encoder.layer.6.attention.k_proj",
+      "model.vision_tower.encoder.layer.6.attention.v_proj",
+      "model.vision_tower.encoder.layer.6.attention.projection_layer",
+      "model.vision_tower.encoder.layer.6.mlp.fc1",
+      "model.vision_tower.encoder.layer.6.mlp.fc2",
+      "model.vision_tower.encoder.layer.7.attention.q_proj",
+      "model.vision_tower.encoder.layer.7.attention.k_proj",
+      "model.vision_tower.encoder.layer.7.attention.v_proj",
+      "model.vision_tower.encoder.layer.7.attention.projection_layer",
+      "model.vision_tower.encoder.layer.7.mlp.fc1",
+      "model.vision_tower.encoder.layer.7.mlp.fc2",
+      "model.vision_tower.encoder.layer.8.attention.q_proj",
+      "model.vision_tower.encoder.layer.8.attention.k_proj",
+      "model.vision_tower.encoder.layer.8.attention.v_proj",
+      "model.vision_tower.encoder.layer.8.attention.projection_layer",
+      "model.vision_tower.encoder.layer.8.mlp.fc1",
+      "model.vision_tower.encoder.layer.8.mlp.fc2",
+      "model.vision_tower.encoder.layer.9.attention.q_proj",
+      "model.vision_tower.encoder.layer.9.attention.k_proj",
+      "model.vision_tower.encoder.layer.9.attention.v_proj",
+      "model.vision_tower.encoder.layer.9.attention.projection_layer",
+      "model.vision_tower.encoder.layer.9.mlp.fc1",
+      "model.vision_tower.encoder.layer.9.mlp.fc2",
+      "model.vision_tower.encoder.layer.10.attention.q_proj",
+      "model.vision_tower.encoder.layer.10.attention.k_proj",
+      "model.vision_tower.encoder.layer.10.attention.v_proj",
+      "model.vision_tower.encoder.layer.10.attention.projection_layer",
+      "model.vision_tower.encoder.layer.10.mlp.fc1",
+      "model.vision_tower.encoder.layer.10.mlp.fc2",
+      "model.vision_tower.encoder.layer.11.attention.q_proj",
+      "model.vision_tower.encoder.layer.11.attention.k_proj",
+      "model.vision_tower.encoder.layer.11.attention.v_proj",
+      "model.vision_tower.encoder.layer.11.attention.projection_layer",
+      "model.vision_tower.encoder.layer.11.mlp.fc1",
+      "model.vision_tower.encoder.layer.11.mlp.fc2",
+      "model.vision_tower.encoder.layer.12.attention.q_proj",
+      "model.vision_tower.encoder.layer.12.attention.k_proj",
+      "model.vision_tower.encoder.layer.12.attention.v_proj",
+      "model.vision_tower.encoder.layer.12.attention.projection_layer",
+      "model.vision_tower.encoder.layer.12.mlp.fc1",
+      "model.vision_tower.encoder.layer.12.mlp.fc2",
+      "model.vision_tower.encoder.layer.13.attention.q_proj",
+      "model.vision_tower.encoder.layer.13.attention.k_proj",
+      "model.vision_tower.encoder.layer.13.attention.v_proj",
+      "model.vision_tower.encoder.layer.13.attention.projection_layer",
+      "model.vision_tower.encoder.layer.13.mlp.fc1",
+      "model.vision_tower.encoder.layer.13.mlp.fc2",
+      "model.vision_tower.encoder.layer.14.attention.q_proj",
+      "model.vision_tower.encoder.layer.14.attention.k_proj",
+      "model.vision_tower.encoder.layer.14.attention.v_proj",
+      "model.vision_tower.encoder.layer.14.attention.projection_layer",
+      "model.vision_tower.encoder.layer.14.mlp.fc1",
+      "model.vision_tower.encoder.layer.14.mlp.fc2",
+      "model.vision_tower.encoder.layer.15.attention.q_proj",
+      "model.vision_tower.encoder.layer.15.attention.k_proj",
+      "model.vision_tower.encoder.layer.15.attention.v_proj",
+      "model.vision_tower.encoder.layer.15.attention.projection_layer",
+      "model.vision_tower.encoder.layer.15.mlp.fc1",
+      "model.vision_tower.encoder.layer.15.mlp.fc2",
+      "model.vision_tower.encoder.layer.16.attention.q_proj",
+      "model.vision_tower.encoder.layer.16.attention.k_proj",
+      "model.vision_tower.encoder.layer.16.attention.v_proj",
+      "model.vision_tower.encoder.layer.16.attention.projection_layer",
+      "model.vision_tower.encoder.layer.16.mlp.fc1",
+      "model.vision_tower.encoder.layer.16.mlp.fc2",
+      "model.vision_tower.encoder.layer.17.attention.q_proj",
+      "model.vision_tower.encoder.layer.17.attention.k_proj",
+      "model.vision_tower.encoder.layer.17.attention.v_proj",
+      "model.vision_tower.encoder.layer.17.attention.projection_layer",
+      "model.vision_tower.encoder.layer.17.mlp.fc1",
+      "model.vision_tower.encoder.layer.17.mlp.fc2",
+      "model.vision_tower.encoder.layer.18.attention.q_proj",
+      "model.vision_tower.encoder.layer.18.attention.k_proj",
+      "model.vision_tower.encoder.layer.18.attention.v_proj",
+      "model.vision_tower.encoder.layer.18.attention.projection_layer",
+      "model.vision_tower.encoder.layer.18.mlp.fc1",
+      "model.vision_tower.encoder.layer.18.mlp.fc2",
+      "model.vision_tower.encoder.layer.19.attention.q_proj",
+      "model.vision_tower.encoder.layer.19.attention.k_proj",
+      "model.vision_tower.encoder.layer.19.attention.v_proj",
+      "model.vision_tower.encoder.layer.19.attention.projection_layer",
+      "model.vision_tower.encoder.layer.19.mlp.fc1",
+      "model.vision_tower.encoder.layer.19.mlp.fc2",
+      "model.vision_tower.encoder.layer.20.attention.q_proj",
+      "model.vision_tower.encoder.layer.20.attention.k_proj",
+      "model.vision_tower.encoder.layer.20.attention.v_proj",
+      "model.vision_tower.encoder.layer.20.attention.projection_layer",
+      "model.vision_tower.encoder.layer.20.mlp.fc1",
+      "model.vision_tower.encoder.layer.20.mlp.fc2",
+      "model.vision_tower.encoder.layer.21.attention.q_proj",
+      "model.vision_tower.encoder.layer.21.attention.k_proj",
+      "model.vision_tower.encoder.layer.21.attention.v_proj",
+      "model.vision_tower.encoder.layer.21.attention.projection_layer",
+      "model.vision_tower.encoder.layer.21.mlp.fc1",
+      "model.vision_tower.encoder.layer.21.mlp.fc2",
+      "model.vision_tower.encoder.layer.22.attention.q_proj",
+      "model.vision_tower.encoder.layer.22.attention.k_proj",
+      "model.vision_tower.encoder.layer.22.attention.v_proj",
+      "model.vision_tower.encoder.layer.22.attention.projection_layer",
+      "model.vision_tower.encoder.layer.22.mlp.fc1",
+      "model.vision_tower.encoder.layer.22.mlp.fc2",
+      "model.vision_tower.encoder.layer.23.attention.q_proj",
+      "model.vision_tower.encoder.layer.23.attention.k_proj",
+      "model.vision_tower.encoder.layer.23.attention.v_proj",
+      "model.vision_tower.encoder.layer.23.attention.projection_layer",
+      "model.vision_tower.encoder.layer.23.mlp.fc1",
+      "model.vision_tower.encoder.layer.23.mlp.fc2",
+      "model.multi_modal_projector.linear_1",
+      "model.multi_modal_projector.linear_2",
+      "lm_head"
+    ],
+    "kv_cache_scheme": null,
+    "quant_method": "compressed-tensors",
+    "quantization_status": "compressed",
+    "sparsity_config": {},
+    "transform_config": {},
+    "version": "0.17.1"
+  },
+  "text_config": {
+    "_name_or_path": "/root/codespace/checkpoints/Qwen3-4B",
+    "architectures": [
+      "Qwen3ForCausalLM"
+    ],
+    "attention_bias": false,
+    "attention_dropout": 0.0,
+    "bos_token_id": 151643,
+    "debug": false,
+    "dtype": "bfloat16",
+    "eos_token_id": 151645,
+    "ep_size": 1,
+    "head_dim": 128,
+    "hidden_act": "silu",
+    "hidden_size": 2560,
+    "initializer_range": 0.02,
+    "intermediate_size": 9728,
+    "layer_types": [
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention"
+    ],
+    "max_position_embeddings": 40960,
+    "max_window_layers": 36,
+    "micro_forward": false,
+    "model_type": "qwen3",
+    "num_attention_heads": 32,
+    "num_hidden_layers": 36,
+    "num_key_value_heads": 8,
+    "pad_token_id": null,
+    "rms_norm_eps": 1e-06,
+    "rope_parameters": {
+      "rope_theta": 1000000,
+      "rope_type": "default"
+    },
+    "skip_checkpoint": false,
+    "sliding_window": null,
+    "tie_word_embeddings": false,
+    "use_cache": true,
+    "use_deepep": false,
+    "use_sliding_window": false,
+    "vocab_size": 151936
+  },
+  "tie_word_embeddings": false,
+  "transformers_version": "5.10.1",
+  "vision_config": {
+    "architectures": [
+      "InternVisionModel"
+    ],
+    "attention_bias": true,
+    "attention_dropout": 0.0,
+    "dropout": 0.0,
+    "dtype": "bfloat16",
+    "hidden_act": "gelu",
+    "hidden_dropout_prob": 0.0,
+    "hidden_size": 1024,
+    "image_size": [
+      448,
+      448
+    ],
+    "initializer_factor": 0.1,
+    "initializer_range": 1e-10,
+    "intermediate_size": 4096,
+    "layer_norm_eps": 1e-06,
+    "layer_scale_init_value": 0.1,
+    "model_type": "internvl_vision",
+    "norm_type": "layer_norm",
+    "num_attention_heads": 16,
+    "num_channels": 3,
+    "num_hidden_layers": 24,
+    "patch_size": [
+      14,
+      14
+    ],
+    "projection_dropout": 0.0,
+    "use_absolute_position_embeddings": true,
+    "use_mask_token": false,
+    "use_mean_pooling": true,
+    "use_qk_norm": false
+  },
+  "vision_feature_layer": -1,
+  "vision_feature_select_strategy": "default"
+}

examples/chat_image.py

@@ -0,0 +1,73 @@
+#!/usr/bin/env python3
+
+from __future__ import annotations
+
+import argparse
+import base64
+import json
+import mimetypes
+import sys
+import urllib.error
+import urllib.request
+from pathlib import Path
+
+
+def build_data_url(image_path: Path) -> str:
+    mime_type, _ = mimetypes.guess_type(image_path.name)
+    if mime_type is None or not mime_type.startswith("image/"):
+        mime_type = "image/jpeg"
+    encoded = base64.b64encode(image_path.read_bytes()).decode("ascii")
+    return f"data:{mime_type};base64,{encoded}"
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser(description="Send a local image to the vLLM OpenAI API.")
+    parser.add_argument("image", type=Path, help="Path to a local image")
+    parser.add_argument("prompt", nargs="?", default="Describe this image in detail.")
+    parser.add_argument("--api-base", default="http://127.0.0.1:8000/v1")
+    parser.add_argument("--model", default="internvl35-fp8")
+    parser.add_argument("--max-tokens", type=int, default=256)
+    args = parser.parse_args()
+
+    if not args.image.is_file():
+        print(f"Image not found: {args.image}", file=sys.stderr)
+        return 2
+
+    payload = {
+        "model": args.model,
+        "messages": [
+            {
+                "role": "user",
+                "content": [
+                    {"type": "image_url", "image_url": {"url": build_data_url(args.image)}},
+                    {"type": "text", "text": args.prompt},
+                ],
+            }
+        ],
+        "temperature": 0.0,
+        "max_tokens": args.max_tokens,
+    }
+
+    request = urllib.request.Request(
+        f"{args.api_base.rstrip('/')}/chat/completions",
+        data=json.dumps(payload).encode("utf-8"),
+        headers={"Content-Type": "application/json"},
+        method="POST",
+    )
+
+    try:
+        with urllib.request.urlopen(request, timeout=300) as response:
+            result = json.load(response)
+    except urllib.error.HTTPError as exc:
+        print(exc.read().decode("utf-8", errors="replace"), file=sys.stderr)
+        return 1
+    except urllib.error.URLError as exc:
+        print(f"Request failed: {exc}", file=sys.stderr)
+        return 1
+
+    print(result["choices"][0]["message"]["content"])
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

examples/chat_text.sh

@@ -0,0 +1,27 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+
+API_BASE="${API_BASE:-http://127.0.0.1:8000/v1}"
+MODEL="${MODEL:-internvl35-fp8}"
+
+curl -sS "$API_BASE/chat/completions" \
+  -H 'Content-Type: application/json' \
+  -d @- <<JSON | python3 -m json.tool
+{
+  "model": "$MODEL",
+  "messages": [
+    {
+      "role": "user",
+      "content": [
+        {
+          "type": "text",
+          "text": "Which number is larger, 9.11 or 9.8? Explain briefly."
+        }
+      ]
+    }
+  ],
+  "temperature": 0.0,
+  "max_tokens": 96
+}
+JSON

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 151643,
+  "eos_token_id": 151645,
+  "transformers_version": "5.10.1"
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c94c1d649e21e763fb440ca4fbcd685169499f01b6e808c5c3618ac54ef92ee
+size 5834016032

processor_config.json

@@ -0,0 +1,79 @@
+{
+  "image_processor": {
+    "crop_to_patches": false,
+    "data_format": "channels_first",
+    "default_to_square": true,
+    "do_convert_rgb": true,
+    "do_normalize": true,
+    "do_rescale": true,
+    "do_resize": true,
+    "image_mean": [
+      0.485,
+      0.456,
+      0.406
+    ],
+    "image_processor_type": "GotOcr2ImageProcessor",
+    "image_std": [
+      0.229,
+      0.224,
+      0.225
+    ],
+    "max_patches": 12,
+    "min_patches": 1,
+    "resample": 3,
+    "rescale_factor": 0.00392156862745098,
+    "size": {
+      "height": 448,
+      "width": 448
+    }
+  },
+  "image_seq_length": 256,
+  "processor_class": "InternVLProcessor",
+  "video_processor": {
+    "data_format": "channels_first",
+    "default_to_square": true,
+    "do_convert_rgb": true,
+    "do_normalize": true,
+    "do_rescale": true,
+    "do_resize": true,
+    "do_sample_frames": false,
+    "image_mean": [
+      0.48145466,
+      0.4578275,
+      0.40821073
+    ],
+    "image_std": [
+      0.26862954,
+      0.26130258,
+      0.27577711
+    ],
+    "initial_shift": true,
+    "model_valid_processing_keys": [
+      "do_convert_rgb",
+      "do_resize",
+      "size",
+      "size_divisor",
+      "default_to_square",
+      "resample",
+      "do_rescale",
+      "rescale_factor",
+      "do_normalize",
+      "image_mean",
+      "image_std",
+      "do_pad",
+      "do_center_crop",
+      "crop_size",
+      "data_format",
+      "input_data_format",
+      "device"
+    ],
+    "resample": 3,
+    "rescale_factor": 0.00392156862745098,
+    "return_metadata": false,
+    "size": {
+      "height": 384,
+      "width": 384
+    },
+    "video_processor_type": "InternVLVideoProcessor"
+  }
+}

quantization/recipe.py

@@ -0,0 +1,18 @@
+"""Compression recipe used to create this checkpoint.
+
+Run quantization from the original BF16 checkpoint, not from an already
+quantized checkpoint.
+"""
+
+from llmcompressor.modifiers.quantization import QuantizationModifier
+
+
+RECIPE = QuantizationModifier(
+    targets="Linear",
+    scheme="FP8_DYNAMIC",
+    ignore=[
+        "re:.*vision_tower.*",
+        "re:.*multi_modal_projector.*",
+        "re:.*lm_head.*",
+    ],
+)

recipe.yaml

@@ -0,0 +1,7 @@
+default_stage:
+  default_modifiers:
+    QuantizationModifier:
+      targets: [Linear]
+      ignore: ['re:.*vision_tower.*', 're:.*multi_modal_projector.*', 're:.*lm_head.*']
+      scheme: FP8_DYNAMIC
+      bypass_divisibility_checks: false

scripts/check_wsl_runtime.sh

@@ -0,0 +1,89 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+
+failures=0
+
+check_command() {
+  local name="$1"
+  if command -v "$name" >/dev/null 2>&1; then
+    echo "OK   command: $name -> $(command -v "$name")"
+  else
+    echo "FAIL command: $name"
+    failures=$((failures + 1))
+  fi
+}
+
+check_file() {
+  local path="$1"
+  if [[ -f "$path" ]]; then
+    echo "OK   file: $path"
+  else
+    echo "FAIL file: $path"
+    failures=$((failures + 1))
+  fi
+}
+
+echo "===== COMMANDS ====="
+check_command python
+check_command vllm
+check_command gcc
+check_command g++
+check_command ninja
+check_command nvidia-smi
+
+echo
+echo "===== SYSTEM FILES ====="
+check_file /usr/include/python3.12/Python.h
+check_file /usr/lib/wsl/lib/libcuda.so
+
+echo
+echo "===== PYTHON PACKAGES ====="
+python - <<'PY'
+import importlib.metadata
+
+for package in ("vllm", "torch", "transformers", "compressed-tensors", "nvidia-cuda-nvrtc"):
+    try:
+        print(f"OK   {package}: {importlib.metadata.version(package)}")
+    except importlib.metadata.PackageNotFoundError:
+        print(f"MISS {package}")
+PY
+
+echo
+echo "===== NVRTC FILES ====="
+python - <<'PY'
+import site
+from pathlib import Path
+
+found = []
+for site_dir in site.getsitepackages():
+    for path in (Path(site_dir) / "nvidia").glob("cu*/lib/libnvrtc*.so*"):
+        found.append(path)
+
+for path in sorted(found):
+    print(path)
+
+if not found:
+    raise SystemExit("No NVRTC libraries found")
+PY
+
+echo
+echo "===== CUDA DRIVER LINK ====="
+cat >/tmp/internvl_cuda_link_test.cpp <<'CPP'
+extern "C" int cuInit(unsigned int flags);
+int main() { return 0; }
+CPP
+
+LIBRARY_PATH="/usr/lib/wsl/lib:${LIBRARY_PATH:-}" \
+  g++ /tmp/internvl_cuda_link_test.cpp -Wl,--no-as-needed -lcuda \
+  -o /tmp/internvl_cuda_link_test
+
+echo "OK   CUDA driver link"
+
+if [[ "$failures" -ne 0 ]]; then
+  echo "Runtime check failed: $failures required item(s) missing." >&2
+  exit 1
+fi
+
+echo
+echo "WSL RUNTIME CHECK: SUCCESS"

scripts/start_vllm_wsl_rtx3070.sh

@@ -0,0 +1,90 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+
+MODEL_ID="${MODEL_ID:-hsmin92/internvl35-fp8}"
+SERVED_MODEL_NAME="${SERVED_MODEL_NAME:-internvl35-fp8}"
+HOST="${HOST:-127.0.0.1}"
+PORT="${PORT:-8000}"
+MAX_MODEL_LEN="${MAX_MODEL_LEN:-2048}"
+MAX_NUM_SEQS="${MAX_NUM_SEQS:-1}"
+MAX_IMAGES="${MAX_IMAGES:-4}"
+GPU_MEMORY_UTILIZATION="${GPU_MEMORY_UTILIZATION:-0.82}"
+KV_CACHE_MEMORY_BYTES="${KV_CACHE_MEMORY_BYTES:-}"
+
+if [[ -z "${VIRTUAL_ENV:-}" ]]; then
+  echo "ERROR: Activate the vLLM virtual environment first." >&2
+  exit 1
+fi
+
+if ! command -v vllm >/dev/null 2>&1; then
+  echo "ERROR: vllm is not installed in the active environment." >&2
+  exit 1
+fi
+
+if [[ ! -f /usr/lib/wsl/lib/libcuda.so ]]; then
+  echo "ERROR: /usr/lib/wsl/lib/libcuda.so was not found." >&2
+  echo "This script is intended for NVIDIA GPU passthrough in WSL2." >&2
+  exit 1
+fi
+
+NVRTC_LIB_DIR="$({ python - <<'PY'
+import site
+from pathlib import Path
+
+candidates = []
+for site_dir in site.getsitepackages():
+    nvidia_dir = Path(site_dir) / "nvidia"
+    if not nvidia_dir.exists():
+        continue
+    for lib_dir in nvidia_dir.glob("cu*/lib"):
+        if any(lib_dir.glob("libnvrtc-builtins.so*")) and any(lib_dir.glob("libnvrtc.so*")):
+            candidates.append(lib_dir)
+
+if not candidates:
+    raise SystemExit(1)
+
+print(sorted(candidates)[-1])
+PY
+} 2>/dev/null)" || {
+  echo "ERROR: pip-installed NVRTC libraries were not found." >&2
+  exit 1
+}
+
+CUDA_DRIVER_LIB_DIR="/usr/lib/wsl/lib"
+
+# WSL2 compatibility settings validated on RTX 3070.
+export VLLM_USE_V2_MODEL_RUNNER=0
+export VLLM_USE_FLASHINFER_SAMPLER=0
+export TOKENIZERS_PARALLELISM=false
+
+# Compile-time linker path for -lcuda.
+export LIBRARY_PATH="$CUDA_DRIVER_LIB_DIR:${LIBRARY_PATH:-}"
+
+# Runtime paths for the WSL CUDA driver and pip-installed NVRTC.
+export LD_LIBRARY_PATH="$CUDA_DRIVER_LIB_DIR:$NVRTC_LIB_DIR:${LD_LIBRARY_PATH:-}"
+
+ARGS=(
+  "$MODEL_ID"
+  --served-model-name "$SERVED_MODEL_NAME"
+  --host "$HOST"
+  --port "$PORT"
+  --trust-remote-code
+  --dtype half
+  --max-model-len "$MAX_MODEL_LEN"
+  --max-num-seqs "$MAX_NUM_SEQS"
+  --gpu-memory-utilization "$GPU_MEMORY_UTILIZATION"
+  --enforce-eager
+  --limit-mm-per-prompt "{\"image\":$MAX_IMAGES,\"video\":0}"
+)
+
+if [[ -n "$KV_CACHE_MEMORY_BYTES" ]]; then
+  ARGS+=(--kv-cache-memory-bytes "$KV_CACHE_MEMORY_BYTES")
+fi
+
+echo "Model: $MODEL_ID"
+echo "NVRTC: $NVRTC_LIB_DIR"
+echo "GPU memory utilization: $GPU_MEMORY_UTILIZATION"
+echo "KV cache bytes: ${KV_CACHE_MEMORY_BYTES:-automatic}"
+
+exec vllm serve "${ARGS[@]}"

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7b9d18660f656ae5a87df2d5d6ed990e80f292d3473c1a35cae8259a5d28cd67
+size 11424484

tokenizer_config.json

@@ -0,0 +1,26 @@
+{
+  "add_prefix_space": false,
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "context_image_token": "<IMG_CONTEXT>",
+  "end_image_token": "</img>",
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "is_local": true,
+  "local_files_only": true,
+  "model_max_length": 40960,
+  "model_specific_special_tokens": {
+    "context_image_token": "<IMG_CONTEXT>",
+    "end_image_token": "</img>",
+    "start_image_token": "<img>",
+    "video_token": "<video>"
+  },
+  "pad_token": "<|endoftext|>",
+  "processor_class": "InternVLProcessor",
+  "split_special_tokens": false,
+  "start_image_token": "<img>",
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null,
+  "video_token": "<video>"
+}