transformers

c12dfddb - [SAM3] Fix precompute vision_embeds or text_embeds for inference (#42407)