finalf0's picture
Add vision_batch_size to avoid cuda OOM (#4)
ceec226 verified