Upload app.py with huggingface_hub

app.py

@@ -52,8 +52,12 @@ else:
 print("Initializing Llama model...")
 llm = Llama(
     model_path=model_path,
-    n_ctx=2048,
-    n_threads=2  # Optimize for HF CPU Space (2 vCPUs)
+    n_ctx=1024,      # Halved context size for faster history processing on CPU
+    n_threads=2,     # Thread-capped to prevent Docker CPU throttling
+    n_batch=256,     # Optimized batch size for CPU Cache
+    use_mmap=False,  # Eagerly loads model into memory during startup (eliminating lazy disk reads)
+    type_k=7,        # Quantizes KV cache keys to Q8_0 (GGML_TYPE_Q8_0)
+    type_v=7         # Quantizes KV cache values to Q8_0 (GGML_TYPE_Q8_0)
 )
 print("Llama model initialized successfully!")