[v1] support quantization (#10161)

src/llamafactory/v1/utils/packages.py

@@ -21,6 +21,13 @@ from functools import lru_cache
 from typing import TYPE_CHECKING
 
 from packaging import version
+from transformers.utils.versions import require_version
+
+from . import logging
+from .env import is_env_enabled
+
+
+logger = logging.get_logger(__name__)
 
 
 if TYPE_CHECKING:
@@ -41,3 +48,22 @@ def _get_package_version(name: str) -> "Version":
 @lru_cache
 def is_transformers_version_greater_than(content: str):
     return _get_package_version("transformers") >= version.parse(content)
+
+
+def check_version(requirement: str, mandatory: bool = False) -> None:
+    r"""Optionally check the package version."""
+    if is_env_enabled("DISABLE_VERSION_CHECK") and not mandatory:
+        logger.warning_rank0_once("Version checking has been disabled, may lead to unexpected behaviors.")
+        return
+
+    if "gptqmodel" in requirement or "autoawq" in requirement:
+        pip_command = f"pip install {requirement} --no-build-isolation"
+    else:
+        pip_command = f"pip install {requirement}"
+
+    if mandatory:
+        hint = f"To fix: run `{pip_command}`."
+    else:
+        hint = f"To fix: run `{pip_command}` or set `DISABLE_VERSION_CHECK=1` to skip this check."
+
+    require_version(requirement, hint)