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

This commit is contained in:
Mick
2025-07-09 23:48:35 -07:00
committed by GitHub
parent 4ed57807c2
commit b5e3d6031c
42 changed files with 887 additions and 524 deletions
@@ -39,7 +39,7 @@ class KimiVLImageProcessor(SGLangBaseProcessor):
max_req_input_len=max_req_input_len,
)
mm_items, input_ids = self.process_and_combine_mm_data(base_output)
mm_items, input_ids, _ = self.process_and_combine_mm_data(base_output)
return {
"input_ids": input_ids.tolist(),