devinxx/nl2bash-combined

# nl2bash-combined

A merged and reformatted dataset for natural language to bash command generation, prepared for fine-tuning instruction-following language models.

## Sources

| Dataset | Split | Rows |
|---|---|---|
| [jiacheng-ye/nl2bash](https://huggingface.co/datasets/jiacheng-ye/nl2bash) | train | 8,090 |
| [jiacheng-ye/nl2bash](https://huggingface.co/datasets/jiacheng-ye/nl2bash) | validation | 609 |
| [jiacheng-ye/nl2bash](https://huggingface.co/datasets/jiacheng-ye/nl2bash) | test | 606 |
| [AnishJoshi/nl2bash-custom](https://huggingface.co/datasets/AnishJoshi/nl2bash-custom) | train | 19,658 |
| [AnishJoshi/nl2bash-custom](https://huggingface.co/datasets/AnishJoshi/nl2bash-custom) | validation | 2,457 |
| [AnishJoshi/nl2bash-custom](https://huggingface.co/datasets/AnishJoshi/nl2bash-custom) | test | 2,458 |

**Total: 33,878 examples** — train: 27,748 / valid: 3,066 / test: 3,064

## Format

Each example is a chat messages list compatible with `mlx-lm` and the Llama/ChatML chat template:

```json
{
"messages": [
{"role": "system", "content": "You are a bash expert. Convert the user's natural language description into a single bash command. Output only the bash command, no explanation."},
{"role": "user", "content": "Find all python files modified in the last 7 days"},
{"role": "assistant", "content": "find . -name '*.py' -mtime -7"}
]
}
```

## Usage

```python
from datasets import load_dataset

ds = load_dataset("devinxx/nl2bash-combined")
print(ds["train"][0])
```

### With mlx-lm fine-tuning

Point your `lora_config.yaml` at this dataset:

```yaml
model: mlx-community/Llama-3.2-1B-Instruct-4bit
data: devinxx/nl2bash-combined
```

## Fine-tuned Model

This dataset was used to train [devinxx/Llama-3.2-1B-nl2bash](https://huggingface.co/devinxx/Llama-3.2-1B-nl2bash) — a 1B parameter model fine-tuned for natural language to bash command generation using LoRA via `mlx-lm`.

## License

Derived from the original nl2bash corpus by Lin et al. (LREC 2018). See original repositories for license details.