# MergeModelCallback[[trl.experimental.merge_model_callback.MergeModelCallback]]

#### trl.experimental.merge_model_callback.MergeModelCallback[[trl.experimental.merge_model_callback.MergeModelCallback]]

```python
trl.experimental.merge_model_callback.MergeModelCallback(merge_config: MergeConfig | None = None, merge_at_every_checkpoint: bool = False, push_to_hub: bool = False)
```

[源代码](https://github.com/huggingface/trl/blob/v1.18.1/trl/experimental/merge_model_callback.py#L294)

**参数：**

merge_config (`experimental.merge_model_callback.MergeConfig`, *可选*) : 用于合并过程的配置。如果未提供，则使用默认的 `MergeConfig`。

merge_at_every_checkpoint (`bool`, *可选*, 默认为 `False`) : 是否在每次保存检查点时合并模型。

push_to_hub (`bool`，*可选*，默认为 `False`)：是否在合并后将合并后的模型推送到 Hub。

一个 [TrainerCallback](https://huggingface.co/docs/transformers/v5.23.0/en/main_classes/callback#transformers.TrainerCallback)，用于根据合并配置将策略模型（即正在训练的模型）与另一个模型进行合并。

示例：

```python
>>> from trl.experimental.merge_model_callback import MergeConfig, MergeModelCallback

>>> config = MergeConfig()
>>> merge_callback = MergeModelCallback(config)
>>> trainer = DPOTrainer(..., callbacks=[merge_callback])
```

