[BugFix] Fix use of per-request seed with pipeline parallel (#6698)

This commit is contained in:
Nick Hill
2024-07-30 10:40:08 -07:00
committed by GitHub
parent f058403683
commit 5cf9254a9c
21 changed files with 222 additions and 137 deletions
+2 -1
View File
@@ -337,7 +337,8 @@ class CPUModelRunner(ModelRunnerBase[CPUModelInput]):
# just use seq_lens instead.
seq_lens,
self.device,
pin_memory=False)
pin_memory=False,
generators=self.get_generators(finished_requests_ids))
return CPUModelInput(
input_tokens=input_tokens,
input_positions=input_positions,