腾讯混元发布 WeMM-Video-4B-GGUF 开源多模态模型
tencent/WeMM-Video-4B-GGUF
腾讯混元发布 WeMM-Video-4B-GGUF 开源多模态模型,支持文本、图像、音频和视频输入,生成文本响应,包含语言模型、视觉编码器、音频编码器,文件总大小 7.53 GiB,采用 Q4_K_M 精度量化,可在 Hugging Face Hub 下载并使用。
Overview
This repository contains the GGUF Q4_K_M bundle of WeMM-Video-4B for the WeMM native runtime. 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.
Files and Quantization
| File | Role | Size |
|---|---|---|
language.gguf |
Language model, tokenizer, and chat template | 4.90 GiB |
mmproj.gguf |
Vision encoder and projector | 0.63 GiB |
audio.gguf |
Audio encoder and projector | 2.00 GiB |
The Q4_K_M recipe protects selected attention and input/output tensors in BF16. See the precision scope and bundle manifest. Keep all three GGUF files at the same Hub revision.
Quick Start
Build the WeMM native runtime and install huggingface_hub. Run the examples from the WeMM-Video repository root to use the bundled media. The launcher fetches the three matching components from the Hub. A prompt by itself sends a text-only request. Use --video for frames alone and --video-audio for frames and the original soundtrack.
export PATH="$PWD/native/install-cpu/bin:$PATH"
MODEL=tencent/WeMM-Video-4B-GGUF
MEDIA=examples/media
python native/run_hf.py --repo-id "$MODEL" \
--prompt "A 20-minute meeting starts at 09:45. When does it end?" --output text-result.json
python native/run_hf.py --repo-id "$MODEL" --image "$MEDIA/apollo-11-earth.jpg" \
--prompt "Describe the planet in this image." --output image-result.json
python native/run_hf.py --repo-id "$MODEL" --audio "$MEDIA/apollo-11-small-step.wav" \
--prompt "Transcribe the English speech in this recording." --output audio-result.json
python native/run_hf.py --repo-id "$MODEL" --video "$MEDIA/apollo-11-first-steps.mp4" \
--max-frames 16 --prompt "Describe what happens on screen." --output video-result.json
python native/run_hf.py --repo-id "$MODEL" --video-audio "$MEDIA/apollo-11-first-steps.mp4" \
--max-frames 16 --prompt "Describe the scene and summarize the speech." --output video-audio-result.json
Choose new --output filenames when rerunning the examples. Use --revision to pin the bundle. See the native guide for CUDA offload.
Validation
The WeMM native CPU runtime loaded the bundle and completed generation checks for text, image, audio, and video with a soundtrack.
License
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.
Citation
@misc{wemmvideo2026,
author = {{Tencent}},
title = {{WeMM-Video}},
year = {2026},
url = {https://github.com/Tencent/WeMM-Video}
}
来源: 腾讯混元 · Hugging Face 新模型 · huggingface.co