update files

app.py

@@ -6,11 +6,70 @@ import os
 import gradio as gr
 from safetensors.torch import load_file
 from huggingface_hub import hf_hub_download
+import gspread
+from google.oauth2 import service_account
+import json
+from datetime import datetime
+import uuid
 
 # Local imports
 from lionguard2 import LionGuard2, CATEGORIES
 from utils import get_embeddings
 
+# Google Sheets configuration
+GOOGLE_SHEET_URL = os.environ.get("GOOGLE_SHEET_URL")
+GOOGLE_CREDENTIALS = os.environ.get("GCP_SERVICE_ACCOUNT")
+RESULTS_SHEET_NAME = "results"
+VOTES_SHEET_NAME = "votes"
+
+# Helper to save results data
+def save_results_data(row):
+    try:
+        # Create credentials object
+        credentials = service_account.Credentials.from_service_account_info(
+            json.loads(GOOGLE_CREDENTIALS),
+            scopes=[
+                "https://www.googleapis.com/auth/spreadsheets",
+                "https://www.googleapis.com/auth/drive",
+            ],
+        )
+        
+        # Create authorized client
+        gc = gspread.authorize(credentials)
+        sheet = gc.open_by_url(GOOGLE_SHEET_URL)
+        ws = sheet.worksheet(RESULTS_SHEET_NAME)
+        ws.append_row(list(row.values()))
+        print(f"Saved results data for text_id: {row['text_id']}")
+    except Exception as e:
+        print(f"Error saving results data: {e}")
+
+# Helper to save vote data
+def save_vote_data(text_id, agree):
+    try:
+        # Create credentials object
+        credentials = service_account.Credentials.from_service_account_info(
+            json.loads(GOOGLE_CREDENTIALS),
+            scopes=[
+                "https://www.googleapis.com/auth/spreadsheets",
+                "https://www.googleapis.com/auth/drive",
+            ],
+        )
+        
+        # Create authorized client
+        gc = gspread.authorize(credentials)
+        sheet = gc.open_by_url(GOOGLE_SHEET_URL)
+        ws = sheet.worksheet(VOTES_SHEET_NAME)
+        
+        vote_row = {
+            "datetime": datetime.now().isoformat(),
+            "text_id": text_id,
+            "agree": agree
+        }
+        ws.append_row(list(vote_row.values()))
+        print(f"Saved vote data for text_id: {text_id}, agree: {agree}")
+    except Exception as e:
+        print(f"Error saving vote data: {e}")
+
 
 def download_model(repo_id, filename="LionGuard2.safetensors", token=None):
     """
@@ -125,13 +184,18 @@ def analyze_text(text):
         
     Returns:
         binary_score: Overall safety score with styling
-        category_table: HTML table with category-specific scores and styling
+        category_table: HTML table with category-specific scores and styling  
+        text_id: Unique identifier for this analysis
+        voting_section: HTML for voting buttons
     """
     if not text.strip():
         empty_html = '<div style="text-align: center; color: #9ca3af; padding: 30px; font-style: italic;">Enter text to analyze</div>'
-        return '<div style="text-align: center; color: #9ca3af; padding: 30px; font-style: italic;">Enter text to analyze</div>', empty_html
+        return '<div style="text-align: center; color: #9ca3af; padding: 30px; font-style: italic;">Enter text to analyze</div>', empty_html, "", ""
     
     try:
+        # Generate unique text ID
+        text_id = str(uuid.uuid4())
+        
         # Get embeddings for the text
         embeddings = get_embeddings([text])
         
@@ -141,6 +205,49 @@ def analyze_text(text):
         # Extract binary score (overall safety)
         binary_score = results.get('binary', [0.0])[0]
         
+        # Extract specific scores for Google Sheets
+        hateful_scores = CATEGORIES['hateful']
+        hateful_l1_score = results.get(hateful_scores[0], [0.0])[0] if len(hateful_scores) > 0 else 0.0
+        hateful_l2_score = results.get(hateful_scores[1], [0.0])[0] if len(hateful_scores) > 1 else 0.0
+        
+        insults_scores = CATEGORIES['insults'] 
+        insults_score = results.get(insults_scores[0], [0.0])[0] if len(insults_scores) > 0 else 0.0
+        
+        sexual_scores = CATEGORIES['sexual']
+        sexual_l1_score = results.get(sexual_scores[0], [0.0])[0] if len(sexual_scores) > 0 else 0.0
+        sexual_l2_score = results.get(sexual_scores[1], [0.0])[0] if len(sexual_scores) > 1 else 0.0
+        
+        physical_violence_scores = CATEGORIES['physical_violence']
+        physical_violence_score = results.get(physical_violence_scores[0], [0.0])[0] if len(physical_violence_scores) > 0 else 0.0
+        
+        self_harm_scores = CATEGORIES['self_harm']
+        self_harm_l1_score = results.get(self_harm_scores[0], [0.0])[0] if len(self_harm_scores) > 0 else 0.0
+        self_harm_l2_score = results.get(self_harm_scores[1], [0.0])[0] if len(self_harm_scores) > 1 else 0.0
+        
+        aom_scores = CATEGORIES['all_other_misconduct']
+        aom_l1_score = results.get(aom_scores[0], [0.0])[0] if len(aom_scores) > 0 else 0.0
+        aom_l2_score = results.get(aom_scores[1], [0.0])[0] if len(aom_scores) > 1 else 0.0
+        
+        # Save results to Google Sheets
+        if GOOGLE_SHEET_URL and GOOGLE_CREDENTIALS:
+            results_row = {
+                "datetime": datetime.now().isoformat(),
+                "text_id": text_id,
+                "text": text,
+                "binary_score": binary_score,
+                "hateful_l1_score": hateful_l1_score,
+                "hateful_l2_score": hateful_l2_score,
+                "insults_score": insults_score,
+                "sexual_l1_score": sexual_l1_score,
+                "sexual_l2_score": sexual_l2_score,
+                "physical_violence_score": physical_violence_score,
+                "self_harm_l1_score": self_harm_l1_score,
+                "self_harm_l2_score": self_harm_l2_score,
+                "aom_l1_score": aom_l1_score,
+                "aom_l2_score": aom_l2_score
+            }
+            save_results_data(results_row)
+        
         # Prepare category data with max scores and dropdowns
         categories_html = []
         
@@ -201,13 +308,38 @@ def analyze_text(text):
         </div>
         '''
         
-        return format_binary_score(binary_score), html_table
+        # Create voting section
+        voting_html = f'''
+        <div style="margin: 24px 0; text-align: center;">
+            <div style="background: #1f2937; border-radius: 12px; padding: 20px; box-shadow: 0 4px 12px rgba(0,0,0,0.3); border: 1px solid #374151;">
+                <h3 style="color: #f9fafb; font-size: 18px; font-weight: 600; margin-bottom: 12px;">📊 How accurate are these results?</h3>
+                <p style="color: #d1d5db; font-size: 14px; margin-bottom: 16px;">Your feedback helps improve the model</p>
+            </div>
+        </div>
+        '''
+        
+        return format_binary_score(binary_score), html_table, text_id, voting_html
         
     except Exception as e:
         error_msg = f"Error analyzing text: {str(e)}"
         error_html = f'<div style="background: linear-gradient(135deg, #991b1b 0%, #b91c1c 100%); color: #fca5a5; padding: 20px; border-radius: 12px; text-align: center; border: 2px solid #ef4444; box-shadow: 0 4px 12px rgba(0,0,0,0.3);">❌ {error_msg}</div>'
-        return f'<div style="background: linear-gradient(135deg, #991b1b 0%, #b91c1c 100%); color: #fca5a5; padding: 16px; border-radius: 8px; text-align: center; border: 1px solid #ef4444;">❌ {error_msg}</div>', error_html
+        return f'<div style="background: linear-gradient(135deg, #991b1b 0%, #b91c1c 100%); color: #fca5a5; padding: 16px; border-radius: 8px; text-align: center; border: 1px solid #ef4444;">❌ {error_msg}</div>', error_html, "", ""
+
+
+# Voting functions
+def vote_thumbs_up(text_id):
+    """Handle thumbs up vote"""
+    if text_id and GOOGLE_SHEET_URL and GOOGLE_CREDENTIALS:
+        save_vote_data(text_id, True)
+        return '<div style="background: linear-gradient(135deg, #065f46 0%, #047857 100%); color: #34d399; padding: 16px; border-radius: 8px; text-align: center; font-weight: 600; border: 1px solid #10b981; margin: 8px 0;">👍 Thanks for your feedback!</div>'
+    return '<div style="color: #9ca3af; text-align: center; padding: 16px;">Voting not available</div>'
 
+def vote_thumbs_down(text_id):
+    """Handle thumbs down vote"""
+    if text_id and GOOGLE_SHEET_URL and GOOGLE_CREDENTIALS:
+        save_vote_data(text_id, False)
+        return '<div style="background: linear-gradient(135deg, #991b1b 0%, #b91c1c 100%); color: #fca5a5; padding: 16px; border-radius: 8px; text-align: center; font-weight: 600; border: 1px solid #ef4444; margin: 8px 0;">👎 Thanks for your feedback!</div>'
+    return '<div style="color: #9ca3af; text-align: center; padding: 16px;">Voting not available</div>'
 
 # Create Gradio interface with dark theme
 with gr.Blocks(title="LionGuard2", theme=gr.themes.Base().set(
@@ -250,6 +382,27 @@ with gr.Blocks(title="LionGuard2", theme=gr.themes.Base().set(
             category_table = gr.HTML(
                 value='<div style="text-align: center; color: #9ca3af; padding: 30px; font-style: italic;">Category scores will appear here after analysis</div>'
             )
+            
+            # Voting section
+            voting_section = gr.HTML(value="")
+            
+            with gr.Row(visible=False) as voting_row:
+                with gr.Column():
+                    gr.HTML("""
+                    <div style="margin: 16px 0; text-align: center;">
+                        <h3 style="color: #f9fafb; font-size: 18px; font-weight: 600; margin-bottom: 8px;">📊 How accurate are these results?</h3>
+                        <p style="color: #d1d5db; font-size: 14px; margin-bottom: 16px;">Your feedback helps improve the model</p>
+                    </div>
+                    """)
+                    
+                    with gr.Row():
+                        thumbs_up_btn = gr.Button("👍 Accurate", variant="secondary", scale=1)
+                        thumbs_down_btn = gr.Button("👎 Inaccurate", variant="secondary", scale=1)
+                    
+                    vote_feedback = gr.HTML(value="")
+            
+            # Hidden text_id to track current analysis
+            current_text_id = gr.Textbox(value="", visible=False)
     
     # Add information about the categories
     with gr.Row():
@@ -275,18 +428,38 @@ with gr.Blocks(title="LionGuard2", theme=gr.themes.Base().set(
             </div>
             """)
     
+    # Function to handle analysis and show voting buttons
+    def analyze_and_show_voting(text):
+        binary_score, category_table, text_id, voting_html = analyze_text(text)
+        # Show voting row if we have results
+        voting_row_update = gr.update(visible=bool(text_id))
+        return binary_score, category_table, text_id, voting_row_update, ""
+    
     # Connect the analyze button to the function
     analyze_btn.click(
-        fn=analyze_text,
+        fn=analyze_and_show_voting,
         inputs=[text_input],
-        outputs=[binary_output, category_table]
+        outputs=[binary_output, category_table, current_text_id, voting_row, vote_feedback]
     )
     
     # Allow Enter key to trigger analysis
     text_input.submit(
-        fn=analyze_text,
+        fn=analyze_and_show_voting,
         inputs=[text_input],
-        outputs=[binary_output, category_table]
+        outputs=[binary_output, category_table, current_text_id, voting_row, vote_feedback]
+    )
+    
+    # Connect voting buttons
+    thumbs_up_btn.click(
+        fn=vote_thumbs_up,
+        inputs=[current_text_id],
+        outputs=[vote_feedback]
+    )
+    
+    thumbs_down_btn.click(
+        fn=vote_thumbs_down,
+        inputs=[current_text_id],
+        outputs=[vote_feedback]
     )
 
 

requirements.txt

@@ -3,4 +3,6 @@ torch>=2.0.0
 safetensors>=0.4.0
 openai>=1.0.0
 numpy>=1.24.0
-huggingface_hub>=0.19.0
+huggingface_hub>=0.19.0
+gspread==6.2.1
+google-auth==2.40.3