[BugFix] Fix use of per-request seed with pipeline parallel (#6698)
This commit is contained in:
@@ -1029,7 +1029,6 @@ class Scheduler:
|
||||
token_chunk_size=token_chunk_size,
|
||||
lora_request=seq_group.lora_request,
|
||||
computed_block_nums=common_computed_block_nums,
|
||||
state=seq_group.state,
|
||||
# `multi_modal_data` will only be present for the 1st comm
|
||||
# between engine and worker.
|
||||
# the subsequent comms can still use delta, but
|
||||
|
||||
Reference in New Issue
Block a user