Update app.py

app.py

@@ -65,7 +65,7 @@ async def lifespan(app: FastAPI):
     llm = Llama(
         model_path=model_path,
         n_gpu_layers=0,             # 0 for CPU execution
-        n_ctx=4096,
+        n_ctx=2048,
         
         # Threading optimizations
         n_threads=4,       
@@ -75,8 +75,6 @@ async def lifespan(app: FastAPI):
         # ⚡ TURBOQUANT EQUIVALENT SETTINGS:
         # Quantizing the KV cache to Q4_0 shrinks memory overhead and drastically
         # reduces RAM bandwidth usage, supercharging CPU generation speeds.
-        type_k=getattr(llama_cpp, "GGML_TYPE_Q4_0", 2),
-        type_v=getattr(llama_cpp, "GGML_TYPE_Q4_0", 2),
         
         n_batch=512,    
         n_ubatch=512,   
@@ -89,7 +87,7 @@ async def lifespan(app: FastAPI):
         use_mmap=True,
         
         # FlashAttention removed as requested
-        flash_attn=True,
+        
     )
     print("Server ready!")