---
license: apache-2.0
configs:
- config_name: RuC-cve2-32k
  data_files:
    - split: train
      path: RuC-cve2-32k.jsonl

- config_name: RuC-tt07-32k
  data_files:
    - split: train
      path: RuC-tt07-32k.jsonl
---


# RuC Datasets

## Summary

RuC is a grammar-driven, rule-selectable benchmark generator that automatically produces RTL code-completion tasks from input hardware description sources. It uses the target HDL grammar to mask syntactically defined code regions and prompts a model to regenerate them using the surrounding unmasked code as context.

**Paper**: https://arxiv.org/abs/2604.27780

RuC-datasets is a collection of hardware design datasets processed with the RuC framework and used for the experiments presented in the paper.

The repository currently contains two datasets:

- `RuC-cve2-32k`
- `RuC-tt07-32k`

Although both datasets share the same schema and preprocessing pipeline, they originate from different source corpora and should be treated as separate datasets.

## Dataset Structure

Each dataset is distributed as a JSONL file where each row corresponds to a single hardware design sample.

Each sample contains the following fields:

- `sample_i`
  Unique sample identifier.

- `topmodule`
  Name of the top-level Verilog/SystemVerilog module.

- `code`
  Full HDL source code of the design.

- `mask_idx`
  Indices of the rule occurrences selected by RuC for masking.

- `all_mask_idx`
  Complete set of candidate rule occurrences eligible for masking.


## Dataset Origins
### RuC-tt07-32k

Derived from the NotSoTiny shuttle tt07 dataset:

https://huggingface.co/datasets/HPAI-BSC/NotSoTiny-25-12

Designs were filtered to retain samples containing fewer than 32000 tokens.

The original source files are Verilog (.v) designs.

### RuC-cve2-32k

Derived from the CVE2 RISC-V core maintained by OpenHWGroup.

The CVE2 codebase was preprocessed using vppreproc and filtered to retain samples containing fewer than 32000 tokens.

The original source files are SystemVerilog (.sv) designs.

## RuC usage

These datasets are intended to be used together with the RuC framework: https://github.com/HPAI-BSC/RuC


To use the samples directly with the RuC framework, each JSON sample must be reconstructed into the original directory layout expected by RuC.

For each sample:
```
RuC-cve2-32k
<sample_i>/
├── <topmodule>.sv
├── mask_idx.json
└── all_mask_idx.json
```

```
RuC-tt07-32k
<sample_i>/
├── <topmodule>.v
├── mask_idx.json
└── all_mask_idx.json
```

Where:
- `<topmodule>.sv` or `<topmodule>.v` contains the contents of the code field.
- `mask_idx.json` contains the contents of the `mask_idx` field.
- `all_mask_idx.json` contains the contents of the `all_mask_idx` field.

The required HDL extension depends on the dataset:
- .sv for `RuC-cve2-32k`
- .v for `RuC-tt07-32k`


## Additional Information
### License
The dataset is released under the Apache License 2.0.

### Citation Information
```
@misc{domingo2026ruchdlagnosticrulecompletion,
      title={RuC: HDL-Agnostic Rule Completion Benchmark Generation}, 
      author={Arnau Ayguadé Domingo and Miquel Alberti-Binimelis and Cristian Gutierrez-Gomez and Emanuele Parisi and Razine Moundir Ghorab and Miquel Moreto and Gokcen Kestor and Dario Garcia-Gasulla},
      year={2026},
      eprint={2604.27780},
      archivePrefix={arXiv},
      primaryClass={cs.AR},
      url={https://arxiv.org/abs/2604.27780}, 
}
```