Fix push to hub logic (#1888)

Author: tastelikefeet

docs/source/LLM/命令行参数.md

@@ -110,7 +110,7 @@
 - `--hub_model_id`: 推送到的ModelScope Hub的model_id, 默认为`None`, 即设置为`f'{model_type}-{sft_type}'`. 你可以将其设置为model_id, 也可以设置为repo_name. 我们会根据hub_token推断出user_name. 推送的远程仓库如果不存在, 则会创建一个新的仓库, 如果存在, 则复用之前的仓库. 该参数只有在`push_to_hub`设置为True时才生效.
 - `--hub_token`: 推送时需要的SDK token. 可以从[https://modelscope.cn/my/myaccesstoken](https://modelscope.cn/my/myaccesstoken)获取, 默认为`None`, 即从环境变量`MODELSCOPE_API_TOKEN`中获取. 该参数只有在`push_to_hub`设置为True时才生效.
 - `--hub_private_repo`: 推送的ModelScope Hub中的模型仓库的权限是否设置为私有, 默认为`False`. 该参数只有在`push_to_hub`设置为True时才生效.
-- `--push_hub_strategy`: 推送策略, 默认为`'push_best'`. 可选择的值包括: 'end', 'push_best', 'push_last', 'checkpoint', 'all_checkpoints'. 'push_best'表示在每次保存权重时, 将最好的模型进行推送并覆盖之前的权重, 'push_last'表示在每次保存权重时, 将最后的权重进行推送并覆盖之前的权重, 'end'表示只在训练的最后推送最好的模型. 该参数只有在`push_to_hub`设置为True时才生效.
+- `--hub_strategy`: 推送策略, 默认为`'every_save'`. 可选择的值包括: 'end', 'every_save', 'checkpoint', 'all_checkpoints'. 该参数从transformers透传而来, 只有在`push_to_hub`设置为True时才生效.
 - `--test_oom_error`: 用于检测训练是否会发生OOM, 默认为`False`. 如果设置为True, 则会将训练集按max_length倒序进行排列, 方便OOM的测试. 该参数一般用于测试, 请谨慎设置.
 - `--disable_tqdm`: 是否不启用tqdm, 这在`nohup`启动脚本时很有用. 默认为`False`, 即为启动tqdm.
 - `--lazy_tokenize`: 如果设置为False,  则在`trainer.train()`之前提前对所有文本进行预处理. 如果设置为True, 则延迟对文本进行编码, 减少预处理的等待并减少内存占用, 这在处理大数据集时很有用. 默认为`None`, 即我们会根据template的类型进行智能选择, LLM的模型通常设置为False, 多模态的模型通常设置为True(避免图片和音频加载导致过多的内存占用).

docs/source_en/LLM/Command-line-parameters.md

@@ -111,7 +111,7 @@
 - `--hub_model_id`: Model_id to push to on ModelScope Hub, default is `None`, i.e. set to `f'{model_type}-{sft_type}'`. You can set this to model_id or repo_name. We will infer user_name based on hub_token. If the remote repository to push to does not exist, a new repository will be created, otherwise the previous repository will be reused. This parameter only takes effect when `push_to_hub` is set to True.
 - `--hub_token`: SDK token needed for pushing. Can be obtained from [https://modelscope.cn/my/myaccesstoken](https://modelscope.cn/my/myaccesstoken), default is `None`, i.e. obtained from environment variable `MODELSCOPE_API_TOKEN`. This parameter only takes effect when `push_to_hub` is set to True.
 - `--hub_private_repo`: Whether to set the permission of the pushed model repository on ModelScope Hub to private, default is `False`. This parameter only takes effect when `push__to_hub` is set to True.
-- `--push_hub_strategy`: Push strategy, default is `'push_best'`. Options include: 'end', 'push_best', 'push_last', 'checkpoint', 'all_checkpoints'. 'push_best' means when saving weights each time, push and overwrite the best model from before, 'push_last' means when saving weights each time, push and overwrite the last weights from before, 'end' means only push the best model at the end of training. This parameter only takes effect when `push_to_hub` is set to True.
+- `--hub_strategy`: Push strategy, default is `'every_save'`. Options include: 'end', 'every_save', 'checkpoint', 'all_checkpoints'. This parameter shares the same meaning from transformers, and only takes effect when `push_to_hub` is set to True.
 - `--test_oom_error`: Used to detect whether training will cause OOM, default is `False`. If set to True, will sort the training set in descending order by max_length, easy for OOM testing. This parameter is generally used for testing, use carefully.
 - `--disable_tqdm`: Whether to disable tqdm, useful when launching script with `nohup`. Default is `False`, i.e. enable tqdm.
 - `--lazy_tokenize`: If set to False, preprocess all text before `trainer.train()`. If set to True, delay encoding text, reducing preprocessing wait and memory usage, useful when processing large datasets. Default is `None`, i.e. we intelligently choose based on template type, usually set to False for LLM models, set to True for multimodal models (to avoid excessive memory usage from loading images and audio).

swift/llm/utils/argument.py

@@ -893,7 +893,7 @@ class SftArguments(ArgumentsBase):
     custom_train_dataset_path: List[str] = field(default_factory=list)
     custom_val_dataset_path: List[str] = field(default_factory=list)
     device_map_config_path: Optional[str] = None
-    push_hub_strategy: Literal['end', 'push_best', 'push_last', 'checkpoint', 'all_checkpoints'] = 'push_best'
+    push_hub_strategy: Optional[Literal['end', 'push_best', 'push_last', 'checkpoint', 'all_checkpoints']] = None
 
     def _prepare_target_modules(self, target_modules) -> Union[List[str], str]:
         if isinstance(target_modules, str):
@@ -1219,7 +1219,7 @@ def _init_training_args(self) -> None:
             adam_epsilon=self.adam_epsilon,
             hub_model_id=self.hub_model_id,
             hub_private_repo=self.hub_private_repo,
-            hub_strategy=self.push_hub_strategy,
+            hub_strategy=self.hub_strategy,
             hub_token=self.hub_token,
             push_to_hub=self.push_to_hub,
             resume_from_checkpoint=self.resume_from_checkpoint,