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

5cf9254a · Nick Hill · GitHub · f0584036 · 5cf9254a
Unverified Commit 5cf9254a authored Jul 30, 2024 by Nick Hill Committed by GitHub Jul 30, 2024
Hide whitespace changes
Inline Side-by-side

Showing with 2 additions and 1 deletion

vllm/worker/xpu_model_runner.py vllm/worker/xpu_model_runner.py +2 -1

No files found.
--- a/vllm/worker/xpu_model_runner.py
+++ b/vllm/worker/xpu_model_runner.py
@@ -246,7 +246,8 @@ class XPUModelRunner(ModelRunnerBase[ModelInputForXPU]):
                # just use seq_lens instead.
                seq_lens,
                self.device,
-                pin_memory=False)
+                pin_memory=False,
+                generators=self.get_generators(finished_requests_ids))
            # Broadcast the metadata.
            metadata_dict = {
                "input_tokens": input_tokens,