Update src/txagent/txagent.py

src/txagent/txagent.py

@@ -78,8 +78,8 @@ class TxAgent:
             model=self.model_name,
             dtype="float16",
             max_model_len=131072,
-            max_num_batched_tokens=32768,  # Increased for A100 80GB
-            gpu_memory_utilization=0.9,    # Higher utilization for better performance
+            max_num_batched_tokens=65536,  # Increased for A100 80GB
+            gpu_memory_utilization=0.95,    # Higher utilization for better performance
             trust_remote_code=True
         )
         self.chat_template = Template(self.model.get_tokenizer().chat_template)
@@ -428,7 +428,7 @@ class TxAgent:
 
         logits_processor = self.build_logits_processor(messages, model)
         sampling_params = SamplingParams(
-            temperature=temperature,
+            temperature=temperature if temperature is not None else 0.0,
             max_tokens=max_new_tokens,
             seed=seed if seed is not None else self.seed,
         )