Update app.py

app.py

@@ -14,7 +14,7 @@ SYSTEM = (
     "No macros, no calories, no conversational text, no markdown, only valid JSON."
 )
 
-llm = Llama(model_path=MODEL_PATH, n_ctx=2048, n_threads=os.cpu_count(), chat_format="chatml")
+llm = Llama(model_path=MODEL_PATH, n_ctx=2048, n_threads=2, chat_format="chatml")
 app = FastAPI()
 
 class MealIn(BaseModel):
@@ -40,7 +40,7 @@ def parse(body: MealIn, authorization: str = Header(default="")):
     out = llm.create_chat_completion(
         messages=[{"role": "system", "content": SYSTEM},
                   {"role": "user", "content": body.meal}],
-        temperature=0, max_tokens=512,
+        temperature=0, max_tokens=256, stop=["<|im_end|>"],
     )
     parsed = extract_json(out["choices"][0]["message"]["content"])
     if parsed is None: