fix #4398 #4592

README.md

@@ -384,6 +384,8 @@ Remember to use `ASCEND_RT_VISIBLE_DEVICES` instead of `CUDA_VISIBLE_DEVICES` to
 
 If you cannot infer model on NPU devices, try setting `do_sample: false` in the configurations.
 
+Download the pre-built Docker images: [32GB](http://mirrors.cn-central-221.ovaijisuan.com/detail/130.html) | [64GB](http://mirrors.cn-central-221.ovaijisuan.com/detail/131.html)
+
 </details>
 
 ### Data Preparation

README_zh.md

@@ -357,7 +357,7 @@ pip install https://github.com/jllllll/bitsandbytes-windows-webui/releases/downl
 
 <details><summary>昇腾 NPU 用户指南</summary>
 
-在昇腾 NPU 设备上安装 LLaMA Factory 时，需要指定额外依赖项，使用 `pip install -e ".[torch-npu,metrics]"` 命令安装。此外，还需要安装 **[Ascend CANN Toolkit and Kernels](https://www.hiascend.com/developer/download/community/result?module=cann)**，安装方法请参考[安装教程](https://www.hiascend.com/document/detail/zh/CANNCommunityEdition/80RC2alpha002/quickstart/quickstart/quickstart_18_0004.html)或使用以下命令：
+在昇腾 NPU 设备上安装 LLaMA Factory 时，需要指定额外依赖项，使用 `pip install -e ".[torch-npu,metrics]"` 命令安装。此外，还需要安装 **[Ascend CANN Toolkit 与 Kernels](https://www.hiascend.com/developer/download/community/result?module=cann)**，安装方法请参考[安装教程](https://www.hiascend.com/document/detail/zh/CANNCommunityEdition/80RC2alpha002/quickstart/quickstart/quickstart_18_0004.html)或使用以下命令：
 
 ```bash
 # 请替换 URL 为 CANN 版本和设备型号对应的 URL
@@ -384,6 +384,8 @@ source /usr/local/Ascend/ascend-toolkit/set_env.sh
 
 如果遇到无法正常推理的情况，请尝试设置 `do_sample: false`。
 
+下载预构建 Docker 镜像：[32GB](http://mirrors.cn-central-221.ovaijisuan.com/detail/130.html) | [64GB](http://mirrors.cn-central-221.ovaijisuan.com/detail/131.html)
+
 </details>
 
 ### 数据准备

src/llamafactory/extras/misc.py

@@ -20,7 +20,9 @@ import os
 from typing import TYPE_CHECKING, Tuple
 
 import torch
+import transformers.dynamic_module_utils
 from transformers import InfNanRemoveLogitsProcessor, LogitsProcessorList
+from transformers.dynamic_module_utils import get_relative_imports
 from transformers.utils import (
     is_torch_bf16_gpu_available,
     is_torch_cuda_available,
@@ -69,6 +71,9 @@ class AverageMeter:
 
 
 def check_dependencies() -> None:
+    r"""
+    Checks the version of the required packages.
+    """
     if os.environ.get("DISABLE_VERSION_CHECK", "0").lower() in ["true", "1"]:
         logger.warning("Version checking has been disabled, may lead to unexpected behaviors.")
     else:
@@ -79,7 +84,7 @@ def check_dependencies() -> None:
         require_version("trl>=0.8.6", "To fix: pip install trl>=0.8.6")
 
 
-def count_parameters(model: torch.nn.Module) -> Tuple[int, int]:
+def count_parameters(model: "torch.nn.Module") -> Tuple[int, int]:
     r"""
     Returns the number of trainable parameters and number of all parameters in the model.
     """
@@ -108,7 +113,7 @@ def count_parameters(model: torch.nn.Module) -> Tuple[int, int]:
     return trainable_params, all_param
 
 
-def get_current_device() -> torch.device:
+def get_current_device() -> "torch.device":
     r"""
     Gets the current available device.
     """
@@ -147,6 +152,13 @@ def get_logits_processor() -> "LogitsProcessorList":
     return logits_processor
 
 
+def has_tokenized_data(path: "os.PathLike") -> bool:
+    r"""
+    Checks if the path has a tokenized dataset.
+    """
+    return os.path.isdir(path) and len(os.listdir(path)) > 0
+
+
 def infer_optim_dtype(model_dtype: "torch.dtype") -> "torch.dtype":
     r"""
     Infers the optimal dtype according to the model_dtype and device compatibility.
@@ -166,11 +178,9 @@ def is_gpu_or_npu_available() -> bool:
     return is_torch_npu_available() or is_torch_cuda_available()
 
 
-def has_tokenized_data(path: "os.PathLike") -> bool:
-    r"""
-    Checks if the path has a tokenized dataset.
-    """
-    return os.path.isdir(path) and len(os.listdir(path)) > 0
+def skip_check_imports() -> None:
+    if os.environ.get("FORCE_CHECK_IMPORTS", "0").lower() not in ["true", "1"]:
+        transformers.dynamic_module_utils.check_imports = get_relative_imports
 
 
 def torch_gc() -> None:

src/llamafactory/model/loader.py

@@ -19,7 +19,7 @@ from transformers import AutoConfig, AutoModelForCausalLM, AutoModelForVision2Se
 from trl import AutoModelForCausalLMWithValueHead
 
 from ..extras.logging import get_logger
-from ..extras.misc import count_parameters, try_download_model_from_ms
+from ..extras.misc import count_parameters, skip_check_imports, try_download_model_from_ms
 from .adapter import init_adapter
 from .model_utils.misc import register_autoclass
 from .model_utils.mod import convert_pretrained_model_to_mod, load_mod_pretrained_model
@@ -48,6 +48,7 @@ def _get_init_kwargs(model_args: "ModelArguments") -> Dict[str, Any]:
 
     Note: including inplace operation of model_args.
     """
+    skip_check_imports()
     model_args.model_name_or_path = try_download_model_from_ms(model_args)
     return {
         "trust_remote_code": True,