[pt/sft] Feature channel loss #4405
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Jintao-Huang
approved these changes
May 29, 2025
| @@ -323,6 +328,11 @@ def __call__( | |||
| if isinstance(dataset_mapped, HfDataset) and len(dataset) != len(dataset_mapped): | |||
| logger.info( | |||
| f'Dataset filtered, origin length: {len(dataset)}, filtered dataset length: {len(dataset_mapped)}') | |||
| if channel: | |||
There was a problem hiding this comment.
这里是为了读取数据的‘channel’字段，把它放到inputs中。
数据示例：{
"channel": "chat",
"messages": [
{"role": "system", "content": "You are a helpful assistant"},
{"role": "user", "content": "What color do you like?"},
{"role": "assistant", "content": "I like blue."}]
}
There was a problem hiding this comment.
这里的map应该是不需要的。dataset_mapped中应该已经包含channel了
There was a problem hiding this comment.
这一步会把channel删掉，是否改成把channel从remove_columns中移除？
swift/trainers/trainers.py
Outdated
| @@ -88,6 +88,8 @@ def __init__(self, *args, **kwargs): | |||
| self.infer_engine = PtEngine.from_model_template( | |||
| self.model, self.template, max_batch_size=self.args.per_device_eval_batch_size) | |||
| self.jsonl_writer = JsonlWriter(os.path.join(self.args.output_dir, 'predict.jsonl')) | |||
| self.channel_cid = None | |||
There was a problem hiding this comment.
这个写个注释吧
channel_cid和cid_channel
There was a problem hiding this comment.
已修改为通过参数方式传入，不需要channel_cid和cid_channel了
added 4 commits
May 30, 2025 11:18
# Conflicts: # swift/trainers/trainers.py
Jintao-Huang
reviewed
Jun 3, 2025
docs/source/Instruction/命令行参数.md
Outdated
| @@ -351,6 +351,7 @@ Vera使用`target_modules`, `target_regex`, `modules_to_save`三个参数. | |||
| - check_model: 检查本地模型文件有损坏或修改并给出提示，默认为True。如果是断网环境，请设置为False。 | |||
| - 🔥create_checkpoint_symlink: 额外创建checkpoint软链接，方便书写自动化训练脚本。best_model和last_model的软链接路径分别为f'{output_dir}/best'和f'{output_dir}/last'。 | |||
| - loss_type: loss类型。默认为None，使用模型自带损失函数。 | |||
| - channel_list : 数据集包含的channel列表。默认为None。结合`--loss_type channel_loss`使用，可参考[这里](https://github.com/modelscope/ms-swift/blob/main/examples/train/plugins/channel_loss.sh)。 | |||
There was a problem hiding this comment.
名字请改成channels
这有可能在训练的时候自动记录channel的实现么
There was a problem hiding this comment.
已处理，见48b4aec
没有在训练的时候自动记录的原因是：多卡情况下，需要保持channels一致(否则可能死锁)，自动记录可能会导致不一致，需要通信来同步，增加通信开销
| @@ -360,6 +360,7 @@ Training arguments include the [base arguments](#base-arguments), [Seq2SeqTraine | |||
| - check_model: Check local model files for corruption or modification and give a prompt, default is True. If in an offline environment, please set to False. | |||
| - 🔥create_checkpoint_symlink: Creates additional checkpoint symlinks to facilitate writing automated training scripts. The symlink paths for `best_model` and `last_model` are `f'{output_dir}/best'` and `f'{output_dir}/last'` respectively. | |||
| - loss_type: Type of loss. Defaults to None, which uses the model's built-in loss function. | |||
| - channel_list：List of channels included in the dataset. Defaults to None. Used in conjunction with `--loss_type channel_loss`. Refer to [this example](https://github.com/modelscope/ms-swift/blob/main/examples/train/plugins/channel_loss.sh) for more details. | |||
swift/llm/train/sft.py
Outdated
| @@ -116,6 +116,7 @@ def run(self): | |||
| train_dataset=train_dataset, | |||
| eval_dataset=val_dataset, | |||
| callbacks=self.callbacks, | |||
| channel_list=self.args.channel_list, | |||
There was a problem hiding this comment.
这个写在TrainArgumentsMixin吧，不在trainer的init方法传入了
| @@ -323,6 +328,11 @@ def __call__( | |||
| if isinstance(dataset_mapped, HfDataset) and len(dataset) != len(dataset_mapped): | |||
| logger.info( | |||
| f'Dataset filtered, origin length: {len(dataset)}, filtered dataset length: {len(dataset_mapped)}') | |||
| if channel: | |||
There was a problem hiding this comment.
这里的map应该是不需要的。dataset_mapped中应该已经包含channel了
Jintao-Huang
approved these changes
Jun 4, 2025
|
还有更新不，打算merge了 |
|
没有更新了，merge吧，谢谢 |
This was referenced Jun 4, 2025
Open
tastelikefeet
added a commit
to tastelikefeet/swift
that referenced
this pull request
Jun 5, 2025
* commit '58b45225287864b4981ac9c0f761a421b4a3f8c0': [vlm] fix llm_lora vlm_full (modelscope#4482) fix infer in client (modelscope#4480) [pt/sft] Feature channel loss (modelscope#4405) [megatron] fix val_dataset (modelscope#4478) fix (modelscope#4475) [seq_parallel] fix sp compute_acc (modelscope#4474)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
PR type
PR information
Monitoring the model's performance on different tasks(channels).
Experiment results