feat: 增强模型下载功能，支持 HuggingFace Hub 直接下载 ONNX 格式模型

ccw/src/templates/dashboard-css/31-api-settings.css

@@ -916,7 +916,7 @@ select.cli-input {
 .cli-textarea {
   resize: vertical;
   min-height: 4rem;
-  max-height: 12rem;
+  max-height: 20rem;
   font-family: 'SF Mono', 'Consolas', 'Liberation Mono', monospace;
   font-size: 0.8125rem;
 }
@@ -2681,7 +2681,7 @@ select.cli-input {
   display: flex;
   position: relative;
   min-height: 200px;
-  max-height: 350px;
+  max-height: min(450px, 50vh);
 }
 
 .json-line-numbers {

ccw/src/templates/dashboard-js/views/codexlens-manager.js

@@ -2672,14 +2672,15 @@ async function loadModelList() {
           '<i data-lucide="plus-circle" class="w-3 h-3"></i> Download Custom Model' +
         '</div>' +
         '<div class="flex gap-2">' +
-          '<input type="text" id="customModelInput" placeholder="e.g., BAAI/bge-small-en-v1.5" ' +
+          '<input type="text" id="customModelInput" placeholder="e.g., Xenova/bge-small-en-v1.5" ' +
             'class="flex-1 text-xs px-2 py-1.5 border border-border rounded bg-background focus:border-primary focus:ring-1 focus:ring-primary outline-none" />' +
           '<button onclick="downloadCustomModel()" class="text-xs px-3 py-1.5 bg-primary text-primary-foreground rounded hover:bg-primary/90">' +
             'Download' +
           '</button>' +
         '</div>' +
-        '<div class="text-[10px] text-muted-foreground mt-2">' +
+        '<div class="text-[10px] text-muted-foreground mt-2 space-y-1">' +
-          'Enter any HuggingFace model name compatible with FastEmbed' +
+          '<div><span class="text-amber-500">⚠</span> Only <strong>ONNX-format</strong> models work with FastEmbed (e.g., Xenova/* models)</div>' +
+          '<div>PyTorch models (intfloat/*, sentence-transformers/*) will download but won\'t work with local embedding</div>' +
         '</div>' +
       '</div>';
     } else {

codex-lens/src/codexlens/cli/model_manager.py

@@ -6,6 +6,12 @@ import shutil
 from pathlib import Path
 from typing import Dict, List, Optional
 
+try:
+    from huggingface_hub import snapshot_download, list_repo_files
+    HUGGINGFACE_HUB_AVAILABLE = True
+except ImportError:
+    HUGGINGFACE_HUB_AVAILABLE = False
+
 try:
     from fastembed import TextEmbedding
     FASTEMBED_AVAILABLE = True
@@ -557,64 +563,85 @@ def download_model(profile: str, progress_callback: Optional[callable] = None) -
 
 def download_custom_model(model_name: str, model_type: str = "embedding", progress_callback: Optional[callable] = None) -> Dict[str, any]:
     """Download a custom model by HuggingFace model name.
-    
+
-    This allows users to download any HuggingFace model that is compatible
+    This allows users to download any HuggingFace model directly from
-    with fastembed (TextEmbedding or TextCrossEncoder).
+    HuggingFace Hub. The model will be placed in the standard cache
-    
+    directory where it can be discovered by scan_discovered_models().
+
+    Note: Downloaded models may not be directly usable by FastEmbed unless
+    they are in ONNX format. This function is primarily for downloading
+    models that users want to use with other frameworks or custom code.
+
     Args:
-        model_name: Full HuggingFace model name (e.g., "BAAI/bge-small-en-v1.5")
+        model_name: Full HuggingFace model name (e.g., "intfloat/e5-small-v2")
-        model_type: Type of model ("embedding" or "reranker")
+        model_type: Type of model ("embedding" or "reranker") - for metadata only
         progress_callback: Optional callback function to report progress
-    
+
     Returns:
         Result dictionary with success status
     """
-    if model_type == "embedding":
+    if not HUGGINGFACE_HUB_AVAILABLE:
-        if not FASTEMBED_AVAILABLE:
+        return {
-            return {
+            "success": False,
-                "success": False,
+            "error": "huggingface_hub not installed. Install with: pip install huggingface_hub",
-                "error": "fastembed not installed. Install with: pip install codexlens[semantic]",
+        }
-            }
+
-    else:
-        if not RERANKER_AVAILABLE:
-            return {
-                "success": False,
-                "error": "fastembed reranker not available. Install with: pip install fastembed>=0.4.0",
-            }
-    
     # Validate model name format (org/model-name)
     if not model_name or "/" not in model_name:
         return {
             "success": False,
-            "error": "Invalid model name format. Expected: 'org/model-name' (e.g., 'BAAI/bge-small-en-v1.5')",
+            "error": "Invalid model name format. Expected: 'org/model-name' (e.g., 'intfloat/e5-small-v2')",
         }
-    
+
     try:
         cache_dir = get_cache_dir()
-        
+
         if progress_callback:
-            progress_callback(f"Downloading custom model {model_name}...")
+            progress_callback(f"Checking model format for {model_name}...")
-        
+
-        if model_type == "reranker":
+        # Check if model contains ONNX files before downloading
-            # Download reranker model
+        try:
-            reranker = TextCrossEncoder(model_name=model_name, cache_dir=str(cache_dir))
+            files = list_repo_files(repo_id=model_name)
+            has_onnx = any(
+                f.endswith('.onnx') or
+                f.startswith('onnx/') or
+                '/onnx/' in f or
+                f == 'model.onnx'
+                for f in files
+            )
+
+            if not has_onnx:
+                return {
+                    "success": False,
+                    "error": f"Model '{model_name}' does not contain ONNX files. "
+                             f"FastEmbed requires ONNX-format models. "
+                             f"Try Xenova/* versions or check the recommended models list.",
+                    "files_found": len(files),
+                    "suggestion": "Use models from the 'Recommended Models' list, or search for ONNX versions (e.g., Xenova/*).",
+                }
+
             if progress_callback:
-                progress_callback(f"Initializing reranker {model_name}...")
+                progress_callback(f"ONNX format detected. Downloading {model_name}...")
-            list(reranker.rerank("test query", ["test document"]))
+
-        else:
+        except Exception as check_err:
-            # Download embedding model
+            # If we can't check, warn but allow download
-            embedder = TextEmbedding(model_name=model_name, cache_dir=str(cache_dir))
             if progress_callback:
-                progress_callback(f"Initializing {model_name}...")
+                progress_callback(f"Could not verify format, proceeding with download...")
-            list(embedder.embed(["test"]))
+
-        
+        # Use huggingface_hub to download the model
+        # This downloads to the standard HuggingFace cache directory
+        local_path = snapshot_download(
+            repo_id=model_name,
+            cache_dir=str(cache_dir),
+        )
+
         if progress_callback:
-            progress_callback(f"Custom model {model_name} downloaded successfully")
+            progress_callback(f"Model {model_name} downloaded successfully")
-        
+
         # Get cache info
         sanitized_name = f"models--{model_name.replace('/', '--')}"
         model_cache_path = cache_dir / sanitized_name
-        
+
         cache_size = 0
         if model_cache_path.exists():
             total_size = sum(
@@ -623,7 +650,7 @@ def download_custom_model(model_name: str, model_type: str = "embedding", progre
                 if f.is_file()
             )
             cache_size = round(total_size / (1024 * 1024), 1)
-        
+
         return {
             "success": True,
             "result": {
@@ -631,6 +658,8 @@ def download_custom_model(model_name: str, model_type: str = "embedding", progre
                 "model_type": model_type,
                 "cache_size_mb": cache_size,
                 "cache_path": str(model_cache_path),
+                "local_path": local_path,
+                "note": "Model downloaded. Note: Only ONNX-format models are compatible with FastEmbed.",
             },
         }