|
|
--- |
|
|
license: mit |
|
|
dataset_info: |
|
|
features: |
|
|
- name: text |
|
|
dtype: string |
|
|
splits: |
|
|
- name: NanoText |
|
|
num_bytes: 6090436 |
|
|
num_examples: 1203 |
|
|
- name: MiniText |
|
|
num_bytes: 60622575 |
|
|
num_examples: 12382 |
|
|
- name: MidiText |
|
|
num_bytes: 181684879 |
|
|
num_examples: 36368 |
|
|
- name: CoreText |
|
|
num_bytes: 606330424 |
|
|
num_examples: 121414 |
|
|
- name: MegaText |
|
|
num_bytes: 1819500227 |
|
|
num_examples: 364168 |
|
|
download_size: 1627122618 |
|
|
dataset_size: 2674228541 |
|
|
configs: |
|
|
- config_name: default |
|
|
data_files: |
|
|
- split: NanoText |
|
|
path: data/NanoText-* |
|
|
- split: MiniText |
|
|
path: data/MiniText-* |
|
|
- split: MidiText |
|
|
path: data/MidiText-* |
|
|
- split: CoreText |
|
|
path: data/CoreText-* |
|
|
- split: MegaText |
|
|
path: data/MegaText-* |
|
|
--- |
|
|
|
|
|
# OpenNeuro: A Dataset to Compute Brain Score Scaling Laws |
|
|
|
|
|
This repository hosts the splits used to train the 20 language models discussed in the associated paper on brain score scaling laws. Each split provides a progressively larger corpus of text, allowing for systematic experimentation at different scales. Below are the key subsets and their statistics. |
|
|
|
|
|
--- |
|
|
|
|
|
## Subset Details |
|
|
|
|
|
### NanoText |
|
|
- **num_bytes**: 6,090,436 |
|
|
- **num_examples**: 1,203 |
|
|
- **Total words**: 1M |
|
|
- **Average words/example**: 831.6 |
|
|
|
|
|
### MiniText |
|
|
- **num_bytes**: 60,622,575 |
|
|
- **num_examples**: 12,382 |
|
|
- **Total words**: 10M |
|
|
- **Average words/example**: 808.1 |
|
|
|
|
|
### MidiText |
|
|
- **num_bytes**: 181,684,879 |
|
|
- **num_examples**: 36,368 |
|
|
- **Total words**: 30M |
|
|
- **Average words/example**: 824.9 |
|
|
|
|
|
### CoreText |
|
|
- **num_bytes**: 606,330,424 |
|
|
- **num_examples**: 121,414 |
|
|
- **Total words**: 100M |
|
|
- **Average words/example**: 823.6 |
|
|
|
|
|
### MegaText |
|
|
- **num_bytes**: 1,819,500,227 |
|
|
- **num_examples**: 364,168 |
|
|
- **Total words**: 300M |
|
|
- **Average words/example**: 823.8 |
|
|
|
|
|
--- |
|
|
|
|
|
## Usage |
|
|
|
|
|
To load any or all of these subsets in Python, install the [🤗 Datasets library](https://github.com/huggingface/datasets) and use: |
|
|
|
|
|
```python |
|
|
from datasets import load_dataset |
|
|
|
|
|
# Load the entire DatasetDict (all splits) |
|
|
dataset_dict = load_dataset("IParraMartin/OpenNeuro") |
|
|
print(dataset_dict) |
|
|
|
|
|
# Or load a specific subset |
|
|
nano_text = load_dataset("IParraMartin/OpenNeuro", split="NanoText") |
|
|
print(nano_text) |
|
|
|
|
|
|
