Video-Text-to-Text
Transformers
Safetensors
English
Chinese
moss_vl
feature-extraction
Realtime
Streaming
Video-Understanding
Image-Understanding
MOSS-VL
OpenMOSS
multimodal
video
vision-language
custom_code
Instructions to use OpenMOSS-Team/MOSS-VL-Realtime with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use OpenMOSS-Team/MOSS-VL-Realtime with Transformers:
# pip install -U transformers accelerate # Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("OpenMOSS-Team/MOSS-VL-Realtime", trust_remote_code=True, device_map="auto") - Notebooks
- Google Colab
- Kaggle
Commit History
23 days ago
-
per-sample media budgets in processor; per-sample MRoPE position ids for batched vision inputs
Verified signatured1f71a5 -
per-sample media budgets in processor; per-sample MRoPE position ids for batched vision inputs
Verified signature75f7473
24 days ago
-
sync offline system prompts with training data: no_thinking -> no system message, deep_thinking -> <think>/<answer> tags
Verified signatureec3e327
-
Unify modeling with the 0708 standard (transformers 4.57 compatible)
Verified signature2cb8df5
-
Unify MOSS-VL processor and inference preprocessing defaults
Verified signatureUse the shared processor and tokenizer, set image/video pixel budgets to 201326592, and disable vision tail padding with vision_seq_pad_multiple=1. Preserve model implementation and weights.5b97f7a
-
Fix query RoPE when reusing cached vision keys
Verified signatureApply query RoPE on every cross-attention step, including cached-vision steps. Rotate new vision keys only once. Preserve the current NPU, idle-wait and text-cleanup code. No weights, configs, processors or tokenizer files changed. CPU fresh/cached regression passed.1e6a45b
-
Ascend NPU support: device-aware cache release, GIL-friendly idle wait, <=8-D vision processing, BICUBIC declaration (#3)
- Ascend NPU support: device-aware cache release, GIL-friendly idle wait, <=8-D vision processing, BICUBIC declaration (5c67b2b784210bc6f92de1decf76a314540c10f9) - simplify: processing reduced to stock + only the <=8-D device-aware 10-D routing (f418456bcc2fab1b5fc8ba2bf150ba7061eb83bd) - revert: preprocessor_config.json to stock (interpolation field was a no-op) (2fb20b58a952c27bb56c7f34f0b9eb469f596047) - fix: strip mid-string U+FFFD in realtime decode emit (3c04eb4ef1662d7facf2fa50a59f7c9aba3ded13) Co-authored-by: Joiin0392 <Joiin0392@users.noreply.huggingface.co>25e81cb
-
Upload README.md with huggingface_hub
Verified signature06b0676
-
Update README.md
Verified signature3fed18c
-
Update README.md
Verified signatureca541ad
-
Update README.md
Verified signature8af1f0c -
Update README.md
Verified signature2bd81d5 -
Update README.md
Verified signature43c9272 -
Upload benchmark-streaming.png
Verified signatureef13d48 -
Delete assets/benchmark-streaming.png
Verified signature2be9a1b -
Update README
Verified signature686e285 -
Upload MOSS-VL-Realtime-0708 release
Verified signaturef55d7c8 -
initial commit
Verified signatured784ba4