1. 修改了一些bug

2. 做了一些调整，比如启动脚本，支持了tenmsorboard

src/api/model_manage.py

@@ -204,44 +204,67 @@ def get_trained_models():
     logger = logging.getLogger(__name__)
 
     try:
-        # 使用 /app/base/saves 目录（容器内路径）
-        saves_base_path = '/app/base/saves'
-        # 本地开发时的备用路径
-        local_saves_path = os.path.join(PROJECT_ROOT, 'saves')
+        # 多个可能的路径
+        potential_paths = [
+            '/app/base/saves',           # 容器内路径
+            os.path.join(PROJECT_ROOT, 'saves'),  # 本地开发路径
+            os.path.join(os.path.dirname(os.path.dirname(PROJECT_ROOT)), 'YG_FT_Base', 'saves'),  # 上级目录
+        ]
 
-        # 选择存在的路径
-        base_path = saves_base_path if os.path.exists(saves_base_path) else local_saves_path
+        base_path = None
+        for path in potential_paths:
+            logger.info(f"[DEBUG] 检查路径: {path}, exists: {os.path.exists(path)}")
+            if os.path.exists(path):
+                base_path = path
+                break
 
-        logger.info(f"[DEBUG] 已训练模型目录: {base_path}, exists: {os.path.exists(base_path)}")
+        logger.info(f"[DEBUG] 最终使用的路径: {base_path}")
 
         models = []
-        if os.path.exists(base_path):
-            for item in os.listdir(base_path):
-                item_path = os.path.join(base_path, item)
-                if os.path.isdir(item_path):
-                    # 检查是否是模板目录（包含训练方法的子目录）
-                    sub_items = []
-                    if os.path.exists(item_path):
-                        for sub_item in os.listdir(item_path):
-                            sub_path = os.path.join(item_path, sub_item)
-                            if os.path.isdir(sub_path):
-                                # 检查是否包含模型文件（adapter_model.bin 或 pytorch_model.bin 等）
-                                has_model = False
-                                for f in os.listdir(sub_path):
-                                    if f.endswith('.bin') or f.endswith('.safetensors'):
-                                        has_model = True
-                                        break
-                                if has_model:
-                                    sub_items.append({
-                                        'name': sub_item,
-                                        'path': sub_path
-                                    })
+        if base_path and os.path.exists(base_path):
+            logger.info(f"[DEBUG] 遍历目录: {base_path}")
+            try:
+                # 路径结构: /app/base/saves/{train_method}/{model_name}/
+                # train_method: lora, full, qlora, dpo, cpt 等
 
-                    models.append({
-                        'name': item,
-                        'path': item_path,
-                        'train_methods': sub_items
-                    })
+                for train_method in os.listdir(base_path):
+                    train_method_path = os.path.join(base_path, train_method)
+                    if not os.path.isdir(train_method_path):
+                        continue
+
+                    logger.info(f"[DEBUG] 检查训练方法目录: {train_method}")
+                    model_count = 0
+
+                    # 遍历模型文件夹
+                    for model_name in os.listdir(train_method_path):
+                        model_path = os.path.join(train_method_path, model_name)
+                        if not os.path.isdir(model_path):
+                            continue
+
+                        # 检查是否有模型文件
+                        try:
+                            files = os.listdir(model_path)
+                            logger.info(f"[DEBUG] {train_method}/{model_name} 文件: {files[:5]}...")
+                            has_model = any(f.endswith('.bin') or f.endswith('.safetensors') for f in files)
+
+                            if has_model:
+                                logger.info(f"[DEBUG] 找到模型: {train_method}/{model_name}")
+                                models.append({
+                                    'name': model_name,
+                                    'path': model_path,
+                                    'train_methods': [{
+                                        'name': train_method,
+                                        'path': model_path
+                                    }]
+                                })
+                                model_count += 1
+                        except Exception as file_err:
+                            logger.error(f"[DEBUG] 读取 {model_path} 失败: {file_err}")
+
+                    logger.info(f"[DEBUG] {train_method} 找到 {model_count} 个模型")
+
+            except Exception as list_err:
+                logger.error(f"[DEBUG] 遍历目录失败: {list_err}")
 
         logger.info(f"[DEBUG] 找到 {len(models)} 个已训练模型")
 
@@ -249,7 +272,7 @@ def get_trained_models():
             'code': 0,
             'data': {
                 'models': models,
-                'base_path': base_path
+                'base_path': base_path or ''
             }
         })
     except Exception as e: