[CUDA Graph] save cuda graph memory by using next_token_logits_buffer (#8579)
This commit is contained in:
@@ -564,6 +564,7 @@ class TboForwardBatchPreparer:
|
||||
mm_inputs=None,
|
||||
top_logprobs_nums=None,
|
||||
token_ids_logprobs=None,
|
||||
next_token_logits_buffer=None,
|
||||
)
|
||||
)
|
||||
|
||||
|
||||
Reference in New Issue
Block a user