fix vllm

huggingface · Nov 22, 2024 · 20c2d22 · 20c2d22
1 parent c3aa25d
commit 20c2d22
Showing 1 changed file with 0 additions and 2 deletions.
diff --git a/optimum_benchmark/backends/vllm/backend.py b/optimum_benchmark/backends/vllm/backend.py
@@ -124,7 +124,6 @@ def batch_offline_engine_generate(self, inputs: Dict[str, Any], kwargs: Dict[str
                     n=kwargs.get("num_return_sequences"),
                     max_tokens=kwargs.get("max_new_tokens"),
                     min_tokens=kwargs.get("min_new_tokens"),
-                    use_beam_search=kwargs.get("num_beams") > 1,
                     logits_processors=kwargs.get("logits_processors", None),
                 ),
             )
@@ -143,7 +142,6 @@ async def single_online_engine_generate(self, prompt: str, request_id: str, kwar
                 n=kwargs.get("num_return_sequences"),
                 max_tokens=kwargs.get("max_new_tokens"),
                 min_tokens=kwargs.get("min_new_tokens"),
-                use_beam_search=kwargs.get("num_beams") > 1,
                 logits_processors=kwargs.get("logits_processors", None),
             ),
         )