---
dataset_info:
  features:
  - name: relation
    dtype: string
  - name: prompt
    dtype: string
  - name: dataset
    dtype: string
  splits:
  - name: train_0
    num_bytes: 136028669
    num_examples: 397625
  - name: eval_0
    num_bytes: 18697180
    num_examples: 54099
  - name: test_0
    num_bytes: 18875338
    num_examples: 53123
  - name: train_2
    num_bytes: 432394197
    num_examples: 397625
  - name: eval_2
    num_bytes: 58205004
    num_examples: 54099
  - name: test_2
    num_bytes: 59493575
    num_examples: 53123
  - name: train_5
    num_bytes: 859491868
    num_examples: 397625
  - name: eval_5
    num_bytes: 115074295
    num_examples: 54099
  - name: test_5
    num_bytes: 118159632
    num_examples: 53123
  download_size: 856690372
  dataset_size: 1816419758
configs:
- config_name: default
  data_files:
  - split: train_0
    path: data/train_0-*
  - split: eval_0
    path: data/eval_0-*
  - split: test_0
    path: data/test_0-*
  - split: train_2
    path: data/train_2-*
  - split: eval_2
    path: data/eval_2-*
  - split: test_2
    path: data/test_2-*
  - split: train_5
    path: data/train_5-*
  - split: eval_5
    path: data/eval_5-*
  - split: test_5
    path: data/test_5-*
task_categories:
  - text-classification
  - text-generation
tags:
  - relation-extraction
  - instruction-tuning
  - general-domain
  - dataset-mixture
pretty_name: GenTune (General-Domain RE Instruction Mixture)
---


# GenTune (General-Domain RE Instruction Mixture)

**GenTune** is an aggregated **instruction-tuning mixture for general-domain relation
extraction (RE)**. It concatenates seven general-domain RE datasets into a single
prompt/label corpus used to fine-tune small language models in the paper **"Sub-Billion,
Super-Frontier: Fine-Tuned Small Language Models Rival Zero-Shot Frontier LLMs on General and
Literary Relation Extraction"** (Christou & Tsoumakas, 2026)
[arXiv:2606.22606](https://arxiv.org/abs/2606.22606).

It is one of the three training regimes in the paper:

- **GenTune** — general-domain only (this dataset)
- **LitTune** — literary only (`Despina/re_littune`)
- **MixTune** — a domain-balanced mix of the two (`Despina/re_mixtune`)

## Component datasets

GenTune is the concatenation (shuffled) of the training/validation/test portions of:

| Source | Repo |
|---|---|
| TACRED | `Despina/tacred` |
| SemEval-2010 Task 8 | `Despina/semeval2010_task8` |
| CoNLL04 | `Despina/conll04` |
| NYT11 | `Despina/nyt_11` |
| GIDS | `Despina/gids` |
| Re-DocRED | `Despina/re-docred` |
| REBEL (subsampled) | `Despina/rebel-dataset` |

## Dataset structure

Each source example is reshaped into a unified instruction-tuning row. The mixture is
materialized once **per shot setting** (0, 2, 5), and each shot setting has its own train /
eval / test split:

| Split | Shot |
|---|---|
| `train_0`, `eval_0`, `test_0` | 0-shot prompts |
| `train_2`, `eval_2`, `test_2` | 2-shot prompts |
| `train_5`, `eval_5`, `test_5` | 5-shot prompts |

### Columns

| Column | Description |
|---|---|
| `prompt` | The instruction prompt for the example at this shot count. For the 0-shot splits, an `"\nAnswer: "` cue is appended to the prompt. |
| `relation` | The gold relation label (the target completion). |
| `dataset` | The source dataset the row came from (e.g. `conll04`, `tacred`), so the mixture can be stratified by source. |

Pick one shot setting per experiment (e.g. train on `train_2` / evaluate on `test_2`) rather
than mixing splits across shot counts.

## Usage

```python
from datasets import load_dataset

# a full DatasetDict with all shot splits
gentune = load_dataset("Despina/re_gentune")
print(gentune)  # train_0, eval_0, test_0, train_2, ... test_5

# or a single split
train2 = load_dataset("Despina/re_gentune", split="train_2")
print(train2[0]["prompt"], "->", train2[0]["relation"])
```

## Licensing and redistribution

> [!WARNING]
> GenTune includes **TACRED** and **NYT11**, which are derived from **Linguistic Data
> Consortium (LDC)** corpora (LDC newswire / the NYT Annotated Corpus, LDC2008T19). Those
> components are **not freely redistributable**: using this mixture requires the appropriate LDC
> licenses, and this aggregate should not be redistributed publicly as-is. If you need a
> shareable general-domain mixture, rebuild GenTune with `src/build_mixtures.py` while excluding
> the LDC-restricted sources.

Each component also retains its own upstream license (CC BY-SA for DocRED/REBEL, etc.); see the
individual dataset cards under the `Despina/` namespace.

## Citation

**If you use this dataset, please cite our paper (and the underlying source datasets you rely on):**

```bibtex
@article{christou2026subbillion,
  title        = {Sub-Billion, Super-Frontier: Small Language Models Rival
                  Zero-Shot Frontier LLMs on General and Literary Relation Extraction},
  author       = {Christou, Despina and Tsoumakas, Grigorios},
  journal      = {arXiv preprint arXiv:2606.22606},
  year         = {2026},
  url          = {https://arxiv.org/abs/2606.22606}
}
```