chore: 忽略离线部署包，提交安全加固、数据库初始化与文档

- .gitignore: 忽略 docker/offline 离线部署包（镜像/运行时等大文件）
- 安全加固: 新增 compute/api/security.py 及各端安全测试，补充 docs/security-hardening.md
- 数据库: 新增完整初始化 SQL 与 docs/database-config.md
- 数据转换与评测: 修复类型检查、增强校验并补充测试
- Docker 配置与环境变量更新

Co-Authored-By: Claude <noreply@anthropic.com>

compute/engines/llama_factory/adapter.py

@@ -15,27 +15,58 @@ class LlamaFactoryCommand:
 
 
 def _load_dataset_preview(path: Path) -> list[dict[str, Any]]:
+    """Load a preview of JSON/JSONL records from a dataset file.
+
+    Content-sniffs instead of trusting the extension so that BOM-prefixed files,
+    JSONL files containing a single JSON array, and mislabeled extensions all work.
+    """
     if not path.exists():
         return []
-    text = path.read_text(encoding="utf-8", errors="replace").strip()
+    text = path.read_text(encoding="utf-8-sig", errors="replace").strip()
     if not text:
         return []
-    if path.suffix.lower() == ".jsonl":
-        items: list[dict[str, Any]] = []
-        for line in text.splitlines()[:20]:
-            line = line.strip()
-            if not line:
-                continue
-            value = json.loads(line)
-            if isinstance(value, dict):
-                items.append(value)
-        return items
-    value = json.loads(text)
+    try:
+        value = json.loads(text)
+    except json.JSONDecodeError:
+        value = None
     if isinstance(value, list):
         return [item for item in value[:20] if isinstance(item, dict)]
     if isinstance(value, dict):
         return [value]
-    return []
+    items: list[dict[str, Any]] = []
+    for line in text.splitlines()[:20]:
+        line = line.strip()
+        if not line:
+            continue
+        try:
+            parsed = json.loads(line)
+        except json.JSONDecodeError:
+            continue
+        if isinstance(parsed, list):
+            items.extend(item for item in parsed[:20] if isinstance(item, dict))
+        elif isinstance(parsed, dict):
+            items.append(parsed)
+        if len(items) >= 20:
+            break
+    return items[:20]
+
+
+def _required_columns_for(formatting: str, columns: dict[str, Any]) -> list[str]:
+    """Required data columns per dataset format.
+
+    Mirrors LLaMA-Factory's leniency: optional columns (e.g. ``input`` / ``query``
+    in Alpaca) are never required, only fields the format structurally needs.
+    """
+    fmt = str(formatting or "").lower()
+    if fmt == "sharegpt":
+        return [str(columns.get("messages") or "messages")]
+    if fmt in {"dpo", "rm", "kto", "ppo"}:
+        return [str(columns[key]) for key in ("chosen", "rejected") if columns.get(key)]
+    if fmt in {"cpt", "pt", "pretrain"}:
+        return [str(columns.get("prompt") or columns.get("text") or "text")]
+    # alpaca family: prompt (instruction) + response (output) required,
+    # query (input) / history are optional and common to omit in jsonl datasets.
+    return [str(columns[key]) for key in ("prompt", "response") if columns.get(key)]
 
 
 def _validate_dataset_columns(config: dict[str, Any]) -> list[str]:
@@ -51,7 +82,7 @@ def _validate_dataset_columns(config: dict[str, Any]) -> list[str]:
         file_name = item.get("file_name")
         file_names = file_name if isinstance(file_name, list) else [file_name]
         columns = item.get("columns") if isinstance(item.get("columns"), dict) else {}
-        required_columns = [str(value) for value in columns.values() if value]
+        required_columns = _required_columns_for(str(item.get("formatting") or ""), columns)
         for name in file_names:
             if not name:
                 continue