feat: redesign custom GGUF loader with native HF search

- Add gradio-huggingfacehub-search dependency for native HF search
- Replace manual textbox+dropdown with HuggingfaceHubSearch component
- Remove 150+ lines of custom search logic (get_popular_gguf_models, search_gguf_models)
- Simplify UI: 1 search component + 1 file dropdown + load button
- Auto-discover GGUF files when model selected (no manual button needed)
- Update event handlers for new auto-discovery flow
- Better UX matching gguf-my-repo space design

app.py

@@ -18,6 +18,7 @@ from llama_cpp import Llama
 from opencc import OpenCC
 import logging
 from huggingface_hub import list_repo_files, hf_hub_download
+from gradio_huggingfacehub_search import HuggingfaceHubSearch
 
 # Configure logging
 logging.basicConfig(level=logging.INFO)
@@ -28,136 +29,6 @@ llm = None
 converter = None
 current_model_key = None
 
-# Global cache for popular GGUF models (populated on first use)
-_popular_gguf_cache: List[Dict[str, Any]] = []
-_popular_gguf_cache_time: float = 0
-_POPULAR_CACHE_TTL = 3600  # 1 hour cache
-
-
-def get_popular_gguf_models(limit: int = 20) -> List[Dict[str, Any]]:
-    """Dynamically fetch popular GGUF models from HuggingFace Hub.
-    
-    Uses HF Hub API to search for models with 'gguf' tag, sorted by downloads.
-    Cached for 1 hour to avoid repeated API calls.
-    
-    Args:
-        limit: Maximum number of models to return
-        
-    Returns:
-        List of model dicts with repo_id, downloads, tags
-    """
-    global _popular_gguf_cache, _popular_gguf_cache_time
-    
-    # Check cache
-    current_time = time.time()
-    if _popular_gguf_cache and (current_time - _popular_gguf_cache_time) < _POPULAR_CACHE_TTL:
-        return _popular_gguf_cache[:limit]
-    
-    try:
-        from huggingface_hub import list_models
-        
-        # Search for models with 'gguf' tag, sorted by downloads (most popular first)
-        models = list_models(
-            filter="gguf",
-            sort="downloads",
-            direction=-1,  # Descending
-            limit=limit * 2,  # Fetch more to filter
-        )
-        
-        # Process and cache results
-        _popular_gguf_cache = []
-        for model in models:
-            # Skip if no GGUF files (just tagged)
-            if not model.tags or "gguf" not in model.tags:
-                continue
-                
-            # Extract parameter count from tags if available
-            params = "Unknown"
-            for tag in model.tags:
-                if "b" in tag.lower() and any(c.isdigit() for c in tag):
-                    params = tag
-                    break
-            
-            _popular_gguf_cache.append({
-                "repo_id": model.id,
-                "downloads": model.downloads,
-                "tags": [t for t in model.tags if t != "gguf"][:5],  # Top 5 non-gguf tags
-                "params": params,
-            })
-            
-            if len(_popular_gguf_cache) >= limit:
-                break
-        
-        _popular_gguf_cache_time = current_time
-        logger.info(f"Cached {len(_popular_gguf_cache)} popular GGUF models from HF Hub")
-        return _popular_gguf_cache
-        
-    except Exception as e:
-        logger.error(f"Failed to fetch popular GGUF models: {e}")
-        # Return empty list on error
-        return []
-
-
-def search_gguf_models(query: str, limit: int = 10) -> List[Dict[str, Any]]:
-    """Search for GGUF models by query string.
-    
-    Searches popular cached models first, then falls back to HF Hub API.
-    
-    Args:
-        query: Search query (partial repo_id or keywords)
-        limit: Maximum results
-        
-    Returns:
-        List of matching model dicts
-    """
-    if not query or len(query) < 2:
-        return []
-    
-    query_lower = query.lower()
-    
-    # First, search in popular models cache
-    popular = get_popular_gguf_models(limit=50)
-    matches = [m for m in popular if query_lower in m["repo_id"].lower()]
-    
-    # If we have enough matches from cache, return them
-    if len(matches) >= limit:
-        return matches[:limit]
-    
-    # Otherwise, try HF Hub API search
-    try:
-        from huggingface_hub import list_models
-        
-        api_models = list_models(
-            search=query,
-            filter="gguf",
-            sort="downloads",
-            direction=-1,
-            limit=limit,
-        )
-        
-        for model in api_models:
-            if model.id not in [m["repo_id"] for m in matches]:
-                params = "Unknown"
-                for tag in model.tags or []:
-                    if "b" in tag.lower() and any(c.isdigit() for c in tag):
-                        params = tag
-                        break
-                
-                matches.append({
-                    "repo_id": model.id,
-                    "downloads": model.downloads,
-                    "tags": [t for t in (model.tags or []) if t != "gguf"][:5],
-                    "params": params,
-                })
-                
-            if len(matches) >= limit:
-                break
-                
-    except Exception as e:
-        logger.error(f"HF Hub search failed: {e}")
-    
-    return matches[:limit]
-
 
 def parse_quantization(filename: str) -> Optional[str]:
     """Extract quantization level from GGUF filename.
@@ -1669,23 +1540,13 @@ def create_interface():
                     
                     # Custom Model UI (hidden by default, shown when custom_hf selected)
                     with gr.Group(visible=False) as custom_model_group:
-                        gr.HTML('<div class="section-header" style="margin-top: 20px;"><span class="section-icon">🔧</span> Custom HuggingFace Model</div>')
-                        
-                        custom_repo_id = gr.Textbox(
-                            label="HuggingFace Repo ID",
-                            placeholder="e.g., unsloth/DeepSeek-R1-Distill-Qwen-7B-GGUF or type to search...",
-                            info="Type model name to search GGUF models on HF, or paste full repo ID",
-                            interactive=True,
-                        )
+                        gr.HTML('<div class="section-header" style="margin-top: 20px;"><span class="section-icon">🔧</span> Load Custom GGUF Model</div>')
                         
-                        # Search results dropdown (shows matching models from HF)
-                        model_search_results = gr.Dropdown(
-                            label="🔍 Search Results - Select a Model",
-                            choices=[],
-                            value=None,
-                            info="Matching GGUF models from HuggingFace Hub will appear here as you type",
-                            interactive=True,
-                            visible=True,
+                        # NEW: Native HF Hub Search Component
+                        model_search_input = HuggingfaceHubSearch(
+                            label="🔍 Search HuggingFace Models",
+                            placeholder="Type model name to search (e.g., 'llama', 'qwen', 'phi')",
+                            search_type="model",
                         )
                         
                         # Hidden fields to store discovered file data
@@ -1693,17 +1554,16 @@ def create_interface():
                         
                         # File dropdown (populated after repo discovery)
                         custom_file_dropdown = gr.Dropdown(
-                            label="📦 Available GGUF Files - Select Precision",
+                            label="📦 Select GGUF File (Precision)",
                             choices=[],
                             value=None,
-                            info="GGUF files from selected repo (alphabetically sorted)",
+                            info="Available GGUF files will appear after selecting a model above",
                             interactive=True,
                             visible=True,
                         )
                         
                         # Action buttons
                         with gr.Row():
-                            discover_btn = gr.Button("🔍 Discover Files", variant="secondary", size="sm")
                             load_btn = gr.Button("⬇️ Load Selected Model", variant="primary", size="sm")
                         
                         # Status message
@@ -2010,103 +1870,56 @@ def create_interface():
                 )
 
         # ==========================================
-        # DYNAMIC MODEL SEARCH - New Feature
+        # NEW: Auto-Discovery Flow with HuggingfaceHubSearch
         # ==========================================
         
-        def search_models_dynamic(query):
-            """Search for GGUF models as user types."""
-            if not query or len(query) < 2:
-                # Clear search results if query too short
-                return gr.update(choices=[], value=None)
-            
-            # Search popular GGUF models from cache
-            matches = search_gguf_models(query, limit=10)
+        def on_model_selected(repo_id):
+            """Handle model selection from HuggingfaceHubSearch.
             
-            if not matches:
-                # No matches found
-                return gr.update(choices=["No matching models found"], value=None)
-            
-            # Format choices with metadata
-            choices = []
-            for m in matches:
-                repo_id = m["repo_id"]
-                params = m.get("params", "Unknown")
-                downloads = m.get("downloads", 0)
-                # Format downloads
-                if downloads >= 1000000:
-                    dl_str = f"{downloads/1000000:.1f}M"
-                elif downloads >= 1000:
-                    dl_str = f"{downloads/1000:.1f}K"
-                else:
-                    dl_str = str(downloads)
-                
-                display = f"{repo_id} | {params} params | ⬇️ {dl_str}"
-                choices.append((display, repo_id))
-            
-            return gr.update(choices=choices, value=None)
-        
-        # Auto-search as user types (with small delay via change event)
-        custom_repo_id.change(
-            fn=search_models_dynamic,
-            inputs=[custom_repo_id],
-            outputs=[model_search_results],
-        )
-        
-        def on_model_selected_from_search(selected_repo):
-            """Handle when user selects a model from search results."""
-            if not selected_repo or selected_repo == "No matching models found":
+            Automatically discovers GGUF files in the selected repo.
+            """
+            if not repo_id:
                 return (
-                    gr.update(value=""),
                     gr.update(choices=[], value=None),
-                    gr.update(visible=True, value="Please select a model from search results"),
                     [],
+                    gr.update(visible=False),
                 )
             
-            # Auto-discover files for selected repo
-            # First show searching status
+            # Show searching status
             yield (
-                gr.update(value=selected_repo),
-                gr.update(choices=["Searching..."], value=None, interactive=False),
-                gr.update(visible=True, value="🔍 Discovering GGUF files..."),
+                gr.update(choices=["🔍 Searching for GGUF files..."], value=None, interactive=False),
                 [],
+                gr.update(visible=True, value=f"Discovering GGUF files in {repo_id}..."),
             )
             
-            files, error = list_repo_gguf_files(selected_repo)
+            # Discover files
+            files, error = list_repo_gguf_files(repo_id)
             
             if error:
                 yield (
-                    gr.update(value=selected_repo),
                     gr.update(choices=[], value=None, interactive=True),
-                    gr.update(visible=True, value=f"❌ {error}"),
                     [],
+                    gr.update(visible=True, value=f"❌ {error}"),
                 )
             elif not files:
                 yield (
-                    gr.update(value=selected_repo),
                     gr.update(choices=[], value=None, interactive=True),
-                    gr.update(visible=True, value="❌ No GGUF files found in this repository"),
                     [],
+                    gr.update(visible=True, value=f"❌ No GGUF files found in {repo_id}"),
                 )
             else:
+                # Format and show files
                 choices = [format_file_choice(f) for f in files]
                 yield (
-                    gr.update(value=selected_repo),
                     gr.update(choices=choices, value=choices[0] if choices else None, interactive=True),
-                    gr.update(visible=True, value=f"✅ Found {len(files)} GGUF files! Select precision and click 'Load Selected Model'"),
                     files,
+                    gr.update(visible=True, value=f"✅ Found {len(files)} GGUF files! Select precision and click 'Load Model'"),
                 )
         
-        # When user selects from search results, auto-fill repo and discover files
-        model_search_results.change(
-            fn=on_model_selected_from_search,
-            inputs=[model_search_results],
-            outputs=[custom_repo_id, custom_file_dropdown, custom_status, custom_repo_files],
-        )
-        
-        # Manual discover button (kept as backup)
-        discover_btn.click(
-            fn=discover_custom_files,
-            inputs=[custom_repo_id],
+        # When user selects from search, auto-discover files
+        model_search_input.change(
+            fn=on_model_selected,
+            inputs=[model_search_input],
             outputs=[custom_file_dropdown, custom_repo_files, custom_status],
         )
 
@@ -2144,14 +1957,14 @@ def create_interface():
 
         load_btn.click(
             fn=load_custom_model_selected,
-            inputs=[custom_repo_id, custom_file_dropdown, custom_repo_files],
+            inputs=[model_search_input, custom_file_dropdown, custom_repo_files],
             outputs=[custom_status, retry_btn, custom_model_state],
         )
 
         # Retry button - same as load
         retry_btn.click(
             fn=load_custom_model_selected,
-            inputs=[custom_repo_id, custom_file_dropdown, custom_repo_files],
+            inputs=[model_search_input, custom_file_dropdown, custom_repo_files],
             outputs=[custom_status, retry_btn, custom_model_state],
         )
 

requirements.txt

@@ -1,3 +1,4 @@
 gradio>=5.0.0
+gradio-huggingfacehub-search>=0.1.0
 opencc-python-reimplemented>=0.1.7
 huggingface-hub>=0.23.0