Update app/main.py

app/main.py

@@ -290,10 +290,11 @@ from contextlib import asynccontextmanager
 from llama_cpp import Llama
 
 # REPO_ID = "FreedomIntelligence/Apollo-7B-GGUF"
-REPO_ID = "RichardErkhov/FreedomIntelligence_-_Apollo-2B-gguf"
 # MODEL_NAME = "FreedomIntelligence/Apollo-7B"
-MODEL_NAME = "FreedomIntelligence/Apollo-2B"
 # MODEL_FILE = "Apollo-7B.Q4_K_S.gguf"
+
+REPO_ID = "RichardErkhov/FreedomIntelligence_-_Apollo-2B-gguf"
+MODEL_NAME = "FreedomIntelligence/Apollo-2B"
 MODEL_FILE = "Apollo-2B.IQ4_XS.gguf"
 EMBEDDING_MODEL = "Omartificial-Intelligence-Space/GATE-AraBert-v1"
 COLLECTION_NAME = "arabic_rag_collection"
@@ -417,7 +418,7 @@ async def ask(query: Query):
         answer = await asyncio.wait_for(
             # qa_chain.run(prompt, callbacks=[timeout_callback]),
             loop.run_in_executor(None, qa_chain.run, prompt),
-            timeout=60
+            timeout=360
         )
 
         if not answer: