Update main.py

main.py

@@ -6,7 +6,7 @@ import uvicorn
 
 app = FastAPI()
 
-client = InferenceClient("meta-llama/Meta-Llama-3.1-405B-Instruct")
+client = InferenceClient("mistralai/Mistral-Large-Instruct-2407")
 
 class Item(BaseModel):
     prompt: str
@@ -35,7 +35,7 @@ def generate(item: Item):
         seed=42,
     )
     formatted_prompt = format_prompt(f"{item.system_prompt}, {item.prompt}", item.history)
-    stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
+    stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False, token='')
     output = ""
 
     for response in stream: