update vllm params (#448)

ModelCloud · Oct 23, 2024 · af8b371 · af8b371
1 parent cdec2a7
commit af8b371
Showing 1 changed file with 1 addition and 4 deletions.
diff --git a/gptqmodel/utils/vllm.py b/gptqmodel/utils/vllm.py
@@ -24,8 +24,6 @@ def convert_hf_params_to_vllm(hf_params: Dict[str, Any]):
         'top_p': hf_params.get('top_p', 1.0),
         'max_tokens': hf_params.get('max_length', 16),
         'min_tokens': hf_params.get('min_length', 0),
-        'early_stopping': hf_params.get('early_stopping', False),
-        'length_penalty': hf_params.get('length_penalty', 1.0),
         'stop_token_ids': [hf_params.get('eos_token_id'), None],
     }
     return SamplingParams(**params)
@@ -58,8 +56,7 @@ def vllm_generate(
     if not isinstance(sampling_params, SamplingParams):
         hf_params = {key: kwargs[key] for key in [
             'num_return_sequences', 'repetition_penalty', 'temperature',
-            'top_k', 'top_p', 'max_length', 'min_length',
-            'early_stopping', 'length_penalty', 'eos_token_id'
+            'top_k', 'top_p', 'max_length', 'min_length', 'eos_token_id'
         ] if key in kwargs}
         sampling_params = convert_hf_params_to_vllm(hf_params)