[Bugfix] Fix single output condition in output processor (vllm-projec…

…t#7881)
Hyper-Accel · Sep 3, 2024 · 0fbc669 · 0fbc669
1 parent 6e36f4f
commit 0fbc669
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/vllm/engine/output_processor/single_step.py b/vllm/engine/output_processor/single_step.py
@@ -113,7 +113,7 @@ def _process_sequence_group_outputs(self, seq_group: SequenceGroup,
                                         outputs: SequenceGroupOutput,
                                         is_async: bool) -> None:
         sampling_params = seq_group.sampling_params
-        if sampling_params.n == 1 and not sampling_params.use_beam_search:
+        if sampling_params.best_of == 1 and not sampling_params.use_beam_search:
             # only have one output sample
             sample = outputs.samples[0]
             # only have one sequence