SOURCE-REVIEWED MODEL PROFILE
WeMM-Video-9B
Dense bilingual audio-visual understanding model with text responses; accepts text, images, audio and video.

Metadata reviewed 2026-11-20 · Original creator: Tencent
PARAMETERS9B (publisher)
CONTEXTRuntime-dependent; 8,192-token serving example
LICENSEapache-2.0
MODALITYText, image, audio and video to text
What it does
- Chinese and English audio-visual understanding
- Timestamped video frames and original soundtrack input
- Separate patched runtime and quantized artifacts
Local fit
BF16 weights reported as 19.54 GiB; additional runtime and media memory required. Requires the project-specific runtime patches; no installation or memory-fit test performed here.
The documented serving example is not a verified maximum context.
Evidence and limits
Source reviewed; no local inference or benchmark.
Publisher runtime required; not independently tested
License and source
Collection license identifies Apache 2.0; retain original third-party component terms and notices.
Read the original model card ↗
Reviewed source revision: 33c084c098d97adbf174a1586a01dfb86a553f00