transformers
Optimize Qwen2VL vision model by precomputing cos/sin embeds before ViT blocks
#35837
Merged
Merged