vlm: support video as an input modality (#5888)

2025-07-09 23:48:35 -07:00
parent 4ed57807c2
commit b5e3d6031c
42 changed files with 887 additions and 524 deletions
@@ -142,7 +142,7 @@ class LlavaVidForCausalLM(nn.Module):
                )
                image_offsets = [
                    flatten_nested_list(
-                        [item.image_offsets for item in image_inputs[i].mm_items]
+                        [item.offsets for item in image_inputs[i].mm_items]
                    )
                    for i in range(bs)
                    if need_vision[i]