Final 5.10.3

src/app.py

@@ -1148,9 +1148,10 @@ Please log in to access GAIA evaluation features with full inference access.
                     refresh_auth_button = gr.Button("🔄 Refresh Auth Status", variant="secondary", scale=1)
                 
                 unit4_run_button = gr.Button(
-                    "🚀 Run GAIA Evaluation & Submit All Answers", 
+                    "🔒 Login Required for GAIA Evaluation", 
                     variant="primary",
-                    scale=2
+                    scale=2,
+                    interactive=False  # Disabled until login
                 )
                 
                 unit4_status_output = gr.Textbox(
@@ -1269,6 +1270,35 @@ Please log in to access GAIA evaluation features with full inference access.
             
             return status, table, auth_status, csv_update, json_update, summary_update
         
+        def refresh_auth_status(request: gr.Request):
+            """Refresh authentication status display"""
+            profile = getattr(request, 'oauth_profile', None)
+            return format_auth_status(profile)
+        
+        def check_login_state(request: gr.Request):
+            """Check if user is logged in and update UI accordingly"""
+            profile = getattr(request, 'oauth_profile', None)
+            
+            if profile:
+                # User is logged in - return updated auth status
+                auth_status = format_auth_status(profile)
+                # Enable the run button
+                button_update = gr.update(interactive=True, value="🚀 Run GAIA Evaluation & Submit All Answers")
+                return auth_status, button_update
+            else:
+                # User not logged in - show login required message
+                auth_status = format_auth_status(None)
+                # Disable the run button  
+                button_update = gr.update(interactive=False, value="🔒 Login Required for GAIA Evaluation")
+                return auth_status, button_update
+        
+        # Set up automatic login state checking
+        interface.load(
+            fn=check_login_state,
+            outputs=[auth_status_display, unit4_run_button],
+            every=2  # Check every 2 seconds for login state changes
+        )
+        
         unit4_run_button.click(
             fn=handle_evaluation_results,
             inputs=[],  # No inputs needed - profile comes from session
@@ -1276,12 +1306,7 @@ Please log in to access GAIA evaluation features with full inference access.
                     csv_download, json_download, summary_download]
         )
         
-        # Refresh authentication status
-        def refresh_auth_status(request: gr.Request):
-            """Refresh authentication status display"""
-            profile = getattr(request, 'oauth_profile', None)
-            return format_auth_status(profile)
-        
+        # Refresh authentication status manually
         refresh_auth_button.click(
             fn=refresh_auth_status,
             outputs=[auth_status_display]

src/models/qwen_client.py

@@ -55,6 +55,11 @@ class QwenClient:
         self.hf_token = hf_token or os.getenv("HUGGINGFACE_TOKEN") or os.getenv("HF_TOKEN")
         if not self.hf_token:
             logger.warning("No HuggingFace token provided. API access may be limited.")
+        
+        # Initialize cost tracking first
+        self.total_cost = 0.0
+        self.request_count = 0
+        self.budget_limit = 0.10  # $0.10 total budget
             
         # Define model configurations with fallbacks
         self.models = {
@@ -123,11 +128,6 @@ class QwenClient:
         self.langchain_clients = {}
         self._initialize_clients()
         
-        # Cost tracking
-        self.total_cost = 0.0
-        self.request_count = 0
-        self.budget_limit = 0.10  # $0.10 total budget
-        
     def _initialize_clients(self):
         """Initialize HuggingFace clients with fallback support"""
         
@@ -351,6 +351,8 @@ class QwenClient:
                 formatted_prompt = f"Question: {prompt}\nAnswer:"
                 
                 try:
+                    # First attempt: Standard formatted prompt
+                    logger.info(f"Attempting generation with {config.name}...")
                     response_text = client.text_generation(
                         formatted_prompt,
                         max_new_tokens=tokens,
@@ -361,33 +363,59 @@ class QwenClient:
                     
                     if not response_text or not response_text.strip():
                         # Try alternative generation method if first fails
-                        logger.warning(f"Empty response from {config.name} attempt 1, trying alternative...")
+                        logger.warning(f"Empty response from {config.name} attempt 1, trying direct prompt...")
                         response_text = client.text_generation(
                             prompt,
-                            max_new_tokens=min(tokens, 100),  # Smaller token limit
-                            temperature=0.7,  # Higher temperature for more response
-                            return_full_text=False
+                            max_new_tokens=min(tokens, 50),  # Smaller token limit
+                            temperature=0.7,  # Higher temperature
+                            return_full_text=False,
+                            do_sample=True
                         )
                     
                     if not response_text or not response_text.strip():
                         logger.warning(f"Empty response from {config.name} attempt 2, trying simple format...")
                         # Try even simpler format
                         response_text = client.text_generation(
-                            f"Answer: {prompt}",
-                            max_new_tokens=50,
-                            temperature=0.5,
+                            f"Answer this: {prompt}",
+                            max_new_tokens=30,
+                            temperature=0.8,
+                            return_full_text=False,
+                            do_sample=True
+                        )
+                        
+                    if not response_text or not response_text.strip():
+                        # Final attempt with minimal parameters
+                        logger.warning(f"Empty response from {config.name} attempt 3, trying minimal config...")
+                        response_text = client.text_generation(
+                            prompt[:100],  # Truncate prompt
+                            max_new_tokens=20,
                             return_full_text=False
                         )
                         
                     if not response_text or not response_text.strip():
-                        raise ValueError(f"No response received from {config.name} after 3 attempts. Response was: '{response_text}'")
+                        error_msg = f"No response received from {config.name} after 4 attempts. Last response: '{response_text}'"
+                        logger.error(f"❌ {error_msg}")
+                        raise ValueError(error_msg)
                         
                 except Exception as gen_error:
-                    logger.error(f"❌ Text generation failed for {config.name}: {gen_error}")
-                    raise ValueError(f"Text generation error: {gen_error}")
+                    error_details = str(gen_error)
+                    logger.error(f"❌ Text generation failed for {config.name}: {error_details}")
+                    
+                    # Check for specific error types
+                    if "timeout" in error_details.lower():
+                        raise ValueError(f"Timeout error with {config.name}: {error_details}")
+                    elif "rate limit" in error_details.lower() or "429" in error_details:
+                        raise ValueError(f"Rate limit error with {config.name}: {error_details}")
+                    elif "auth" in error_details.lower() or "401" in error_details:
+                        raise ValueError(f"Authentication error with {config.name}: {error_details}")
+                    else:
+                        raise ValueError(f"Generation error with {config.name}: {error_details}")
                 
+                # Final validation
                 if not response_text or not response_text.strip():
-                    raise ValueError(f"Final response check failed for {config.name}. Response: '{response_text}'")
+                    error_msg = f"Final validation failed for {config.name}. Response: '{response_text}'"
+                    logger.error(f"❌ {error_msg}")
+                    raise ValueError(error_msg)
             
             response_time = time.time() - start_time
             

src/production_deployment_guide.md

@@ -90,7 +90,44 @@ After successful deployment with enhanced fallback system:
 - **Smart Error Recovery**: Advanced retry logic with multiple generation attempts
 - **Enhanced Debugging**: Detailed error reporting for troubleshooting
 
-### **📊 Latest Production Fixes (v2.1):**
+### **📊 Latest Production Fixes (v2.2):**
+
+#### Dynamic Authentication Detection ✅
+```python
+# Real-time login state monitoring:
+interface.load(
+    fn=check_login_state,
+    outputs=[auth_status_display, unit4_run_button],
+    every=2  # Check every 2 seconds for login state changes
+)
+
+# Button state updates based on login:
+if profile:
+    button_update = gr.update(interactive=True, value="🚀 Run GAIA Evaluation & Submit All Answers")
+else:
+    button_update = gr.update(interactive=False, value="🔒 Login Required for GAIA Evaluation")
+```
+
+#### Model Initialization Bug Fixes ✅
+```python
+# Fixed QwenClient total_cost attribute error:
+def __init__(self, hf_token: Optional[str] = None):
+    # Initialize cost tracking FIRST
+    self.total_cost = 0.0
+    self.request_count = 0
+    self.budget_limit = 0.10
+    # Then initialize models...
+```
+
+#### Enhanced FLAN-T5 Generation ✅
+```python
+# 4-attempt generation strategy:
+1. Standard formatted prompt: "Question: {prompt}\nAnswer:"
+2. Direct prompt with higher temperature
+3. Simple format: "Answer this: {prompt}"  
+4. Minimal config with truncated prompt
+# Each with detailed error logging and specific error type detection
+```
 
 #### OAuth Scope Request Configuration ✅
 ```python
@@ -108,32 +145,6 @@ login_button = gr.LoginButton(
 )
 ```
 
-#### HF_TOKEN Priority System ✅
-```python
-# Authentication priority order:
-1. HF_TOKEN environment variable (highest priority)
-2. OAuth token from user login (now requests inference access)
-3. SimpleClient fallback (guaranteed to work)
-```
-
-#### Enhanced Error Handling ✅
-```python
-# Multi-attempt generation with detailed logging:
-1. Standard generation with formatted prompt
-2. Alternative generation with different parameters  
-3. Simple format generation as last resort
-4. Graceful fallback to SimpleClient if all fail
-```
-
-#### Guaranteed Fallback Logic ✅
-```python
-# Even with HF_TOKEN, if models fail:
-if main_client_fails:
-    log_detailed_error()
-    fallback_to_simple_client()  # Always works
-    ensure_user_gets_responses()
-```
-
 ## 🎯 Deployment Steps
 
 ### 1. Pre-Deployment Checklist