NVIDIA-NeMo / NVIDIA-NeMo/RL

Unify dataset_name and dataset_cls

Open
#1,614 0 comments 0 reactions 0 assignees View on GitHub
data module enhancement
Dominant language
Python
Stars
2k
Forks
561
Avg merge
4d 5h
Merged PRs (30d)
145

Description

Currently we are using `dataset_name` in data config, but it is not intuitive since that’s sometimes a dataset class.

1. Remove `dataset_name`, always use `dataset_cls` to make it easy to understand.
2. Support `dataset_cls` to load from path to support a custom `dataset_cls`.
3. Keep the compatibility of `dataset_name` for now, and remove all of them later.

```python
# [Before]
# for supported dataset
dataset_name: OpenMathInstruct-2

# for custom dataset
dataset_name: ResponseDataset
train_data_path:

# [After]
# for supported dataset
dataset_cls: OpenMathInstruct2Dataset

# for custom dataset
dataset_cls: ResponseDataset
train_data_path:
```

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.