This repository contains the NVFP4 W4A16 Safetensors variant of WeMM-Video-4B. WeMM-Video is a Chinese–English bilingual family for joint audio-visual understanding and reasoning. It accepts text, images, audio, and video and generates text responses.
WeMM-Video can combine video frames with the original soundtrack to describe events, transcribe speech, and reason about what is seen and heard. It also accepts silent video, standalone audio, images, and text, with responses in Chinese or English.
Vision and audio encoders turn media into tokens that the language model processes alongside text. Video frames carry timestamps. The family was pretrained on approximately 1.2 trillion multimodal tokens, then post-trained with multi-teacher OPD, anti-repetition MrDPO, and specialist reinforcement learning.
See the project overview for family details.
Language feed-forward weights use group-16 NVFP4. Activations use BF16. Other tensors, including attention weights and norms, retain their source precision. See the quantization manifest for tensor scope and export checks.
This checkpoint uses ModelOpt MIXED_PRECISION. Use the patched vLLM and Marlin recipe below for these weights. The corresponding BF16 checkpoint has a direct Transformers loading example.
Install the matching WeMM runtime patch. The serving guide lists the dependency profiles and Docker images. Choose one server below.
Use patched vLLM 0.31.0 with ModelOpt Mixed Precision and Marlin (SM120 reference configuration). See the vLLM recipe for setup details.
MODEL=tencent/WeMM-Video-4B-NVFP4
vllm serve "$MODEL" --served-model-name "$MODEL" --trust-remote-code --dtype bfloat16 \
--host 127.0.0.1 --port 8000 --max-model-len 8192 \
--max-num-seqs 1 --max-num-batched-tokens 8192 \
--no-enable-prefix-caching --mamba-cache-mode none --mamba-ssm-cache-dtype float32 \
--mm-processor-cache-gb 0 --skip-mm-profiling \
--limit-mm-per-prompt '{"image":4,"video":2,"audio":2}' \
--media-io-kwargs '{"video":{"fps":2,"min_frames":4,"max_frames":16,"total_pixels":262144}}' \
--attention-config '{"backend":"FLASH_ATTN","flash_attn_version":2}' \
--mm-encoder-attn-backend FLASH_ATTN \
--quantization modelopt_mixed --linear-backend marlin --moe-backend marlin
With the server running on port 8000, use the bundled request client from the project root in a second terminal. It sends media inline to the OpenAI-compatible chat endpoint.
MODEL=tencent/WeMM-Video-4B-NVFP4
python examples/chat.py text --model "$MODEL"
python examples/chat.py image --model "$MODEL"
python examples/chat.py audio --model "$MODEL"
python examples/chat.py video --model "$MODEL"
python examples/chat.py video-audio --model "$MODEL"
| Input | Request content |
|---|---|
| Text | Text string in messages[].content |
| Image | image_url with image_url.url |
| Audio | audio_url with audio_url.url |
| Video frames | video_url with video_url.use_audio set to false |
| Video with soundtrack | video_url with video_url.use_audio set to true |
Set fps and max_frames inside video_url to control sampling for a clip. Set chat_template_kwargs.enable_thinking to true for Thinking mode. See the complete video request.
The checkpoint template defaults to Instruct. Set enable_thinking=True through the runtime's chat-template options to use Thinking mode. Use the media processor and chat template shipped with this checkpoint.
Patched vLLM 0.31.0 loaded and ran the PRO 5000 reference weights with ModelOpt Mixed Precision and Marlin. The five generation checks covered text, image, audio, silent video, and video with a soundtrack. Validation reused the weights already on the test machine.
See the WeMM-Video-35B-A3B evaluation results for the benchmark comparison in the technical report.
Apache License 2.0. See the included license and the collection license and third-party attribution. Third-party components remain subject to their original terms.
@misc{WeMM-Video,
title={WeMM-Video: Pushing the Frontiers of Open-Source Omni-Modal Video Understanding},
author={Changli Tang and Jie Yang and Tianyi Wang and Lei Li and Daoze Zhang and Zijie Xin and Zhan Liu and Li Li and Jinchang Xu and Fengyun Rao and Chao Zhang and Jing Lyu},
year={2026},
eprint={2611.26409},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2611.26409},
}
tencent/WeMM-Video-4B