Update app.py

app.py

@@ -10,9 +10,9 @@ import requests
 app = Flask(__name__)
 CORS(app)
 
-# --- CORRECTED MODEL ENDPOINT ---
-# We are switching to Google's Gemma model, which is highly reliable.
-LLM_API_URL = "https://api-inference.huggingface.co/models/google/gemma-2b-it"
+# --- FAILSAFE MODEL ENDPOINT ---
+# Using TinyLlama, a reliable and non-gated model.
+LLM_API_URL = "https://api-inference.huggingface.co/models/TinyLlama/TinyLlama-1.1B-Chat-v1.0"
 HF_TOKEN = os.environ.get("HF_TOKEN")
 
 @app.route('/')
@@ -51,12 +51,15 @@ def analyze():
         if not HF_TOKEN:
             return jsonify({"error": "Hugging Face API token is not configured on the server."}), 500
 
-        prompt = f"""<|user|>
+        # This prompt format is optimized for TinyLlama/Zephyr models.
+        prompt = f"""<|system|>
 You are ReflectAI. A person's photo shows an expression of '{attributes['emotion']}', they are a {attributes['gender']} aged around {attributes['age']}.
 Write a two-part response:
 1. **Comment:** A short, insightful, and encouraging comment about their aura.
 2. **Advice:** A brief, actionable piece of advice related to their expression or age.
-Format the output with "Comment:" and "Advice:" labels. Do not repeat the attributes.<|end|>
+Format the output with "Comment:" and "Advice:" labels. Do not repeat the attributes.</s>
+<|user|>
+Generate the response.</s>
 <|assistant|>
 """