Question

[vlm-driver]

Thanks for your nice work!

I'd like to know the evaluation time for LLaMA-Vision-11B.

[safevl]

LLaMA-Vision-11B is approximately 2.3 seconds per sample on an A100 GPU (40GB). This includes the time for both image encoding and text generation, with a maximum generation length of 100 tokens. The actual inference time may vary depending on your specific hardware configuration and the complexity of the input image.