Updated

app.py

@@ -17,22 +17,23 @@ model_path = hf_hub_download(repo_id=MODEL_REPO, filename=MODEL_FILE)
 # For 0.5B model, 512-1024 is the "sweet spot" for free CPU Spaces
 llm_instance = Llama(
     model_path=model_path,
-    n_ctx=512, 
-    n_batch=512,
+    n_ctx=128, 
+    n_batch=128,
     use_mmap=True,
     verbose=False
 )
 
 # --- LLAMA-INDEX WORKER ---
 # This object is ready for your indexing/RAG logic
+'''
 index_llm = LlamaCPP(
     model_path=model_path,
-    context_window=512,
+    context_window=256,
     max_new_tokens=256,
     model_kwargs={"n_gpu_layers": 0}, 
     verbose=False
 )
-
+'''
 class ChatRequest(BaseModel):
     prompt: str
 
@@ -43,7 +44,7 @@ async def chat(request: ChatRequest):
     
     response = llm_instance(
         prompt, 
-        max_tokens=256, 
+        max_tokens=128, 
         stop=["<|im_end|>", "<|endoftext|>"]
     )
     return {"response": response["choices"][0]["text"]}