Update main.py

main.py

@@ -60,7 +60,7 @@ def read_root(request: Request, input_data: InputData):
     generated_response = generate(input_text, history, temperature, max_new_tokens, top_p, repetition_penalty)
     return {"response": generated_response}
 
-def generate(prompt, history, temperature=0.2, max_new_tokens=30000, top_p=0.95, repetition_penalty=1.0):
+def generate(prompt, history, temperature=0.7, max_new_tokens=30000, top_p=0.95, repetition_penalty=1.0):
     temperature = float(temperature)
     if temperature < 1e-2:
         temperature = 1e-2
@@ -116,7 +116,7 @@ async def GeneraTestoAsync(url, input_data):
                 'top_p': input_data.top_p,
                 'repetition_penalty': input_data.repetition_penalty
             }
-            tasks.append(make_request(session, token, data, url) for _ in range(input_data.NumeroGenerazioni))
+            tasks.extend([make_request(session, token, data, url) for _ in range(input_data.NumeroGenerazioni)])
         #tasks = [make_request(session, token, data, url) for _ in range(input_data.NumeroGenerazioni)]  
         return await asyncio.gather(*tasks)