vlm: support video as an input modality (#5888)
This commit is contained in:
@@ -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(),
|
||||
|
||||
Reference in New Issue
Block a user