Update model.py

model.py

@@ -11,7 +11,7 @@ if torch.cuda.is_available():
         model_id,
         torch_dtype=torch.float16,
         device_map='cuda',
-    )
+    ).to("cuda")
 else:
     model = AutoModelForCausalLM.from_pretrained(
         model_id,
@@ -44,7 +44,7 @@ def run(message: str,
         chat_history: list[tuple[str, str]],
         system_prompt: str,
         max_new_tokens: int = 1024,
-        temperature: float = 0.1,
+        temperature: float = 0.2,
         top_p: float = 0.95,
         top_k: int = 50) -> Iterator[str]:
     prompt = get_prompt(message, chat_history, system_prompt)
@@ -62,7 +62,6 @@ def run(message: str,
         top_p=top_p,
         top_k=top_k,
         temperature=temperature,
-        num_beams=1,
         eos_token_id=tokenizer.eos_token_id,
         pad_token_id=tokenizer.pad_token_id,
     )