Update app.py

app.py

@@ -46,7 +46,7 @@ logger = logging.getLogger(__name__)
 # ══════════════════════════════════════════════════════════════════════════════
 
 def get_client():
-    """Spaces-optimized client initialization with debug logging."""
+    """Spaces-optimized client initialization using official HF Inference provider."""
     if not HF_AVAILABLE:
         logger.error("❌ HuggingFace Hub not available")
         return None
@@ -58,7 +58,6 @@ def get_client():
     api_token = os.getenv("HF_API_TOKEN")
     if api_token:
         logger.info(f"✅ Found HF_API_TOKEN (length: {len(api_token)})")
-        logger.info(f"✅ Token starts with: {api_token[:10]}..." if len(api_token) > 10 else "✅ Token found but very short")
     else:
         logger.warning("❌ HF_API_TOKEN not found in environment")
     
@@ -80,10 +79,6 @@ def get_client():
         except Exception as e:
             logger.warning(f"❌ CLI token check failed: {e}")
     
-    # Debug: Show all environment variables that start with HF_
-    hf_env_vars = {k: v[:10] + "..." if v else None for k, v in os.environ.items() if k.startswith('HF_')}
-    logger.info(f"🔍 All HF_ environment variables: {hf_env_vars}")
-    
     if not api_token:
         logger.error("❌ No HF token found in any location")
         return None
@@ -94,23 +89,34 @@ def get_client():
         return None
     
     try:
-        logger.info("🔄 Initializing HuggingFace client for Spaces...")
-        client = InferenceClient(token=api_token)
+        logger.info("🔄 Initializing HuggingFace client with hf-inference provider...")
+        
+        # FIXED: Use the official provider-based approach
+        client = InferenceClient(
+            provider="hf-inference",
+            api_key=api_token,
+        )
         
-        # Simple connectivity test (optional)
+        # Test with official example from documentation
         logger.info("🧪 Testing client connectivity...")
         try:
-            # Try fill-mask with bert-base-uncased (more likely to be available)
-            test_result = client.fill_mask("Paris is the [MASK] of France.", model="bert-base-uncased")
-            logger.info(f"✅ Client test successful with bert-base-uncased - got {len(test_result)} results")
+            # Use official fill_mask example from docs
+            test_result = client.fill_mask(
+                "The answer to the universe is [MASK].",
+                model="google-bert/bert-base-uncased",
+            )
+            logger.info(f"✅ Client test successful - got {len(test_result)} results")
         except Exception as e1:
             try:
-                # Fallback: try text classification (almost always available)
-                test_result = client.text_classification("This is a test.", model="distilbert-base-uncased-finetuned-sst-2-english")
-                logger.info(f"✅ Client test successful with sentiment model")
+                # Fallback test with text classification
+                test_result = client.text_classification(
+                    "I like you. I love you",
+                    model="NousResearch/Minos-v1",
+                )
+                logger.info(f"✅ Client test successful with text classification")
             except Exception as e2:
-                # If tests fail, client is still valid - just skip testing
-                logger.info("✅ Client initialized (skipping model tests - they may be loading)")
+                # If tests fail, client might still work - just log
+                logger.info("✅ Client initialized (model tests may be loading)")
         
         logger.info("✅ HuggingFace client ready for use")
         return client
@@ -689,4 +695,5 @@ if __name__ == "__main__":
         server_port=7860,
         show_error=True,
         quiet=False
-    )
+    )
+