Update app.py

app.py

@@ -13,11 +13,9 @@ OPENAI_API_URL = "https://models.inference.ai.azure.com/chat/completions"
 OPENAI_MODEL_NAME = "gpt-4o-mini"
 
 # 2. Google Gemini Configuration (Direct Google API)
-# You need to set GOOGLE_API_KEY in your HF Space secrets
 GOOGLE_API_KEY = os.environ.get("GOOGLE_API_KEY", "") 
 
-# CORRECTED: The model name is gemma-2-27b-it (Gemma 2), not gemma-3.
-# NOTE: While this uses the "Gemini" API endpoint, it calls the Gemma 2 open model.
+# CORRECTED: Use gemma-2-27b-it (Gemma 2).
 GEMINI_API_URL = f"https://generativelanguage.googleapis.com/v1beta/models/gemma-3-27b-it:generateContent?key={GOOGLE_API_KEY}"
 
 app = FastAPI(
@@ -51,12 +49,8 @@ def home():
 
 @app.get("/check-limit")
 def check_limit():
-    """
-    Checks the rate limit status of OpenAI tokens. 
-    (Google API doesn't provide easy rate limit headers in the same way, skipped for now).
-    """
+    """Checks the rate limit status of OpenAI tokens."""
     if not AI_SERVICE_TOKENS:
-        # Just return empty if no OpenAI tokens, but don't crash if Google is used
         return {"tokens_checked": 0, "results": [], "note": "OpenAI tokens missing"}
 
     results = []
@@ -116,12 +110,12 @@ def call_google_gemini(filename):
     if not GOOGLE_API_KEY:
         raise Exception("GOOGLE_API_KEY not configured.")
     
-    # Construct the Gemini payload
+    # Updated Prompt: Since we can't use JSON mode, we make the prompt stricter.
     prompt = f"""
     You are an expert Movie and TV metadata analyst. 
     Analyze the filename: "{filename}"
     Identify the title, year, and whether it is a series.
-    Return ONLY a raw JSON object with this exact format: 
+    Return ONLY a raw JSON object with this exact format (no markdown, no backticks): 
     {{"title": "Movie Title", "year": "2024", "isSeries": false}}
     """
     
@@ -131,19 +125,17 @@ def call_google_gemini(filename):
         }],
         "generationConfig": {
             "temperature": 0.1,
-            "maxOutputTokens": 100,
-            "responseMimeType": "application/json" # Hints the model to output JSON
+            "maxOutputTokens": 100
+            # REMOVED: "responseMimeType": "application/json" (Not supported by Gemma)
         }
     }
 
-    # Note: The URL here uses the global GEMINI_API_URL defined at the top
     response = requests.post(GEMINI_API_URL, headers={"Content-Type": "application/json"}, json=payload, timeout=30)
     
     if response.status_code != 200:
         raise Exception(f"Google Gemini API Error {response.status_code}: {response.text}")
     
     result = response.json()
-    # Extract text from Gemini response structure
     try:
         return result['candidates'][0]['content']['parts'][0]['text']
     except (KeyError, IndexError):
@@ -159,14 +151,11 @@ def analyze_filename(request: AnalyzeRequest):
 
     try:
         if provider_used == "gemma":
-            # Although the frontend sends "gemma", we map this to our Google Gemini function
             raw_content = call_google_gemini(request.filename)
         else:
-            # Default to OpenAI
             if not AI_SERVICE_TOKENS: raise HTTPException(500, "OpenAI tokens missing.")
             raw_content = call_openai_gpt4o(request.filename, AI_SERVICE_TOKENS)
 
-        # Parse JSON output from either provider
         if raw_content:
             # Clean up markdown code blocks if the model includes them
             clean_content = raw_content.replace("```json", "").replace("```", "").strip()