Update app.py

app.py

@@ -1,7 +1,7 @@
 from fastapi import FastAPI, HTTPException
 from pydantic import BaseModel
 from llama_cpp import Llama
-from concurrent.futures import ThreadPoolExecutor
+from concurrent.futures import ThreadPoolExecutor, as_completed
 import uvicorn
 from dotenv import load_dotenv
 from difflib import SequenceMatcher
@@ -10,7 +10,7 @@ load_dotenv()
 
 app = FastAPI()
 
-# Inicialización de los modelos
+# Configuración de los modelos
 models = [
     {"repo_id": "Ffftdtd5dtft/gpt2-xl-Q2_K-GGUF", "filename": "gpt2-xl-q2_k.gguf"},
     {"repo_id": "Ffftdtd5dtft/Meta-Llama-3.1-8B-Instruct-Q2_K-GGUF", "filename": "meta-llama-3.1-8b-instruct-q2_k.gguf"},
@@ -47,7 +47,7 @@ def select_best_response(responses, request):
     return best_response
 
 def filter_by_coherence(responses, request):
-    # Puedes implementar un filtro más sofisticado si es necesario
+    # Implementa aquí un filtro de coherencia si es necesario
     return responses
 
 def filter_by_similarity(responses):
@@ -62,16 +62,18 @@ def filter_by_similarity(responses):
 
 @app.post("/generate_chat")
 async def generate_chat(request: ChatRequest):
-    with ThreadPoolExecutor() as executor:
-        # Ejecutar las tareas en paralelo
+    # Ejecutar en ThreadPoolExecutor sin límite explícito de workers
+    with ThreadPoolExecutor(max_workers=None) as executor:
         futures = [executor.submit(generate_chat_response, request, llm) for llm in llms]
-        responses = [future.result() for future in futures]
+        responses = []
+        for future in as_completed(futures):
+            response = future.result()
+            responses.append(response)
     
     if any("Error" in response for response in responses):
         error_response = next(response for response in responses if "Error" in response)
         raise HTTPException(status_code=500, detail=error_response)
     
-    # Seleccionar la mejor respuesta
     best_response = select_best_response(responses, request)
     return {"response": best_response}