Modelsslug: qwen3-vl

Qwen3-VL

Alibaba's Qwen vision-language model family encoding visual tokens with spatiotemporal coordinates.

Curated Intelligence

1 items · Page 1/1
量子位 (微信公众号)T2·Models
74 pts
V-CAST Cuts Long-Video Inference Time by Half Without Training

Researchers from Sun Yat-sen University and Hong Kong Polytechnic University propose V-CAST, a training-free, plug-and-play visual token pruning method that allocates per-frame budgets and preserves original spatiotemporal coordinates before tokens enter the language model. On 768-frame audio-visual benchmarks it retains 97.1% of the uncompressed model's performance while cutting total inference time by 50.0%, and on Qwen3-VL-8B it keeps 98.2% average performance using only 25% of visual tokens.