Update main.py

main.py

@@ -3,7 +3,7 @@ from fastapi import FastAPI, Form
 from pydantic import BaseModel
 
 #Model loading
-llm = AutoModelForCausalLM.from_pretrained("TheBloke/Toppy-M-7B-GGUF", model_file="toppy-m-7b.Q2_K.gguf",
+llm = AutoModelForCausalLM.from_pretrained("TheBloke/Toppy-M-7B-GGUF", model_file="toppy-m-7b.Q5_K_S.gguf",
 model_type='mistral',
 max_new_tokens = 1096,
 threads = 3,