moving share link under input field on the sparse reps tab

app.py

@@ -9,56 +9,9 @@ import os
 
 # Add this CSS at the top of your file, after the imports
 css = """
-/* Move share button to top-right corner */
-.share-button {
-    position: fixed !important;
-    top: 20px !important;
-    right: 20px !important;
-    z-index: 1000 !important;
-    background: #4CAF50 !important;
-    color: white !important;
-    border-radius: 8px !important;
-    padding: 8px 16px !important;
-    font-weight: bold !important;
-    box-shadow: 0 2px 10px rgba(0,0,0,0.2) !important;
-}
-
-.share-button:hover {
-    background: #45a049 !important;
-    transform: translateY(-1px) !important;
-}
-
-/* Alternative positions - uncomment the one you want instead */
-
-/* Top-left corner */
-/*
-.share-button {
-    position: fixed !important;
-    top: 20px !important;
-    left: 20px !important;
-    z-index: 1000 !important;
-}
-*/
-
-/* Bottom-right corner (mobile-friendly) */
-/*
-.share-button {
-    position: fixed !important;
-    bottom: 20px !important;
-    right: 20px !important;
-    z-index: 1000 !important;
-}
-*/
-
-/* Bottom-center */
-/*
-.share-button {
-    position: fixed !important;
-    bottom: 20px !important;
-    left: 50% !important;
-    transform: translateX(-50%) !important;
-    z-index: 1000 !important;
-}
+/* The global fixed positioning for the share button is no longer needed
+   because we'll place gr.ShareButton directly in the UI.
+   You can remove or comment out any previous .share-button CSS if it was there.
 */
 """
 
@@ -130,11 +83,11 @@ def create_lexical_bow_mask(input_ids_batch, vocab_size, tokenizer):
                 tokenizer.unk_token_id
             ]:
                 meaningful_token_ids.append(token_id)
-                
+
         if meaningful_token_ids:
             # Apply mask to the current row in the batch
             bow_masks[i, list(set(meaningful_token_ids))] = 1
-            
+
     return bow_masks
 
 
@@ -185,7 +138,7 @@ def get_splade_cocondenser_representation(text):
 
     info_output = f"" # Line 1
     info_output += f"Total non-zero terms in vector: {len(indices)}\n" # Line 2 (and onwards for sparsity)
-    
+
 
     return formatted_output, info_output
 
@@ -243,7 +196,7 @@ def get_splade_lexical_representation(text):
 
     info_output = f"" # Line 1
     info_output += f"Total non-zero terms in vector: {len(indices)}\n" # Line 2 (and onwards for sparsity)
-    
+
 
     return formatted_output, info_output
 
@@ -260,11 +213,11 @@ def get_splade_doc_representation(text):
     binary_bow_vector = create_lexical_bow_mask(
         inputs['input_ids'], vocab_size, tokenizer_splade_doc
     ).squeeze() # Squeeze back for single output
-    
+
     indices = torch.nonzero(binary_bow_vector).squeeze().cpu().tolist()
     if not isinstance(indices, list):
         indices = [indices] if indices else []
-        
+
     values = [1.0] * len(indices) # All values are 1 for binary representation
     token_weights = dict(zip(indices, values))
 
@@ -338,12 +291,12 @@ def get_splade_lexical_vector(text):
             torch.log(1 + torch.relu(output.logits)) * inputs['attention_mask'].unsqueeze(-1),
             dim=1
         )[0].squeeze()
-        
+
         vocab_size = tokenizer_splade_lexical.vocab_size
         bow_mask = create_lexical_bow_mask(
             inputs['input_ids'], vocab_size, tokenizer_splade_lexical
         ).squeeze()
-        
+
         splade_vector = splade_vector * bow_mask
         return splade_vector
     return None
@@ -377,7 +330,7 @@ def format_sparse_vector_output(splade_vector, tokenizer, is_binary=False):
         values = [1.0] * len(indices)
     else:
         values = splade_vector[indices].cpu().tolist()
-        
+
     token_weights = dict(zip(indices, values))
 
     meaningful_tokens = {}
@@ -408,8 +361,8 @@ def format_sparse_vector_output(splade_vector, tokenizer, is_binary=False):
 
     # This is the line that will now always be split into two
     info_output = f"Total non-zero terms: {len(indices)}\n" # Line 1
-    
-    
+
+
     return formatted_output, info_output
 
 
@@ -451,7 +404,7 @@ def calculate_dot_product_and_representations_independent(query_model_choice, do
     # Combine output into a single string for the Markdown component
     full_output = f"### Dot Product Score: {dot_product:.6f}\n\n"
     full_output += "---\n\n"
-    
+
     # Query Representation
     full_output += f"Query Representation ({query_model_name_display}):\n\n"
     full_output += query_main_rep_str + "\n\n" + query_info_str # Added an extra newline for better spacing
@@ -460,7 +413,7 @@ def calculate_dot_product_and_representations_independent(query_model_choice, do
     # Document Representation
     full_output += f"Document Representation ({doc_model_name_display}):\n\n"
     full_output += doc_main_rep_str + "\n\n" + doc_info_str # Added an extra newline for better spacing
-    
+
     return full_output
 
 
@@ -488,7 +441,13 @@ with gr.Blocks(title="SPLADE Demos", css=css) as demo:
                         label="Enter your query or document text here:",
                         placeholder="e.g., Why is Padua the nicest city in Italy?"
                     )
-                    # New Markdown component for the info output
+                    # --- NEW: Place the gr.ShareButton here ---
+                    gr.ShareButton(
+                        value="Share My Sparse Representation",
+                        components=[input_text, model_radio], # You can specify components to share
+                        visible=True # Make sure it's visible
+                    )
+                    # --- End New ---
                     info_output_display = gr.Markdown(
                         value="",
                         label="Vector Information",
@@ -496,7 +455,7 @@ with gr.Blocks(title="SPLADE Demos", css=css) as demo:
                     )
                 with gr.Column(scale=2): # Right column for the main representation output
                     main_representation_output = gr.Markdown()
-            
+
             # Connect the interface elements
             model_radio.change(
                 fn=predict_representation_explorer,
@@ -508,15 +467,16 @@ with gr.Blocks(title="SPLADE Demos", css=css) as demo:
                 inputs=[model_radio, input_text],
                 outputs=[main_representation_output, info_output_display]
             )
-            
+
             # Initial call to populate on load (optional, but good for demo)
             demo.load(
                 fn=lambda: predict_representation_explorer(model_radio.value, input_text.value),
                 outputs=[main_representation_output, info_output_display]
             )
 
-        with gr.TabItem("Compare Encoders"): # NEW TAB
-            
+        with gr.TabItem("Compare Encoders"): # Reverted to original gr.Interface setup
+            gr.Markdown("### Calculate Dot Product Similarity Between Encoded Query and Document")
+
             # Define the common model choices for cleaner code
             model_choices = [
                 "MLM encoder (SPLADE-cocondenser-distil)",
@@ -549,7 +509,7 @@ with gr.Blocks(title="SPLADE Demos", css=css) as demo:
                     )
                 ],
                 outputs=gr.Markdown(),
-                allow_flagging="never"
+                allow_flagging="never" # Keep this to keep the share button at the bottom of THIS interface
             )
 
 demo.launch()