Update main.py

main.py

@@ -7,8 +7,15 @@ from llama_cpp import Llama
 app = FastAPI()
 
 # --- CONFIGURATION ---
-REPO_ID = "CoolShotSystems/Axiom-3.1-Sovereign"
-FILENAME = "Meta-Llama-3.1-8B.Q4_K_M.gguf"
+# REPO_ID = "CoolShotSystems/Axiom-3.1-Sovereign"
+# FILENAME = "Meta-Llama-3.1-8B.Q4_K_M.gguf"
+
+# --- CONFIGURATION ---
+# OLD (Slow): "CoolShotSystems/Axiom-3.1-Sovereign"
+# NEW (Fast): "hugging-quants/Llama-3.2-3B-Instruct-Q4_K_M-GGUF"
+
+REPO_ID = "hugging-quants/Llama-3.2-3B-Instruct-Q4_K_M-GGUF"
+FILENAME = "llama-3.2-3b-instruct-q4_k_m.gguf"
 
 # Global Model Variable
 axiom_model = None