Update app.py

app.py

@@ -23,12 +23,11 @@ model_path = hf_hub_download(
 )
 
 print("Initializing memory-optimized llama.cpp execution runtime...")
-# Hardened configurations specifically tuned to avoid OOM crashes on free shared CPU tiers
 llm = Llama(
     model_path=model_path, 
-    n_ctx=4096,   # Shifting context window to 4K slashes active memory requirements in half
-    n_batch=32,   # Drop processing batches from 512 to 32 to eliminate peak memory spikes
-    n_threads=2   # Explicitly bounds CPU thread contention blocks
+    n_ctx=6144,   # Expanded context window from 4096 to 6144 for longer historical tracks
+    n_batch=32,   # Kept at 32 to guarantee flat peak memory profiles
+    n_threads=2
 )
 
 @app.get("/")
@@ -39,13 +38,12 @@ def read_root():
 async def chat_completions(request: Request):
     body = await request.json()
     messages = body.get("messages", [])
-    
     response = llm.create_chat_completion(
         messages=messages,
         temperature=0.3,
         top_p=0.95,
         stream=True,
-        max_tokens=1536
+        max_tokens=3072  # Doubled token headroom from 1536 to 3072
     )
     
     def stream_generator():