Dubeg

main.py

@@ -17,10 +17,10 @@ def read_root(
         model: str = "meta-llama/Llama-2-7b-hf",
         temperature: float = 0.,
         max_tokens: int = 1024) -> List:
-    sampling_params = SamplingParams(temperature=temperature, max_tokens=max_tokens, token=token)
+    sampling_params = SamplingParams(temperature=temperature, max_tokens=max_tokens)
 
     llm = LLM(model=model)
 
-    response = llm.generate([prompt], sampling_params)
+    response = llm.generate([prompt], sampling_params, token=token)
 
     return response