---
dataset_info:
  features:
  - name: chunk_index
    dtype: int64
  - name: chunk_text
    dtype: string
  - name: chunk_tokens
    sequence: int64
  - name: chunk_token_count
    dtype: int64
  - name: id
    dtype: string
  - name: url
    dtype: string
  - name: score
    dtype: float64
  - name: dump
    dtype: string
  - name: embedding
    sequence: float64
  - name: __index_level_0__
    dtype: int64
  splits:
  - name: train
    num_bytes: 296035820712
    num_examples: 25504378
  download_size: 215649217827
  dataset_size: 296035820712
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
license: apache-2.0
pretty_name: FineWeb-edu 10BT Sample embedded with nomic-text-v1.5
size_categories:
- 10M<n<100M
---
# FineWeb-edu 10BT Sample embedded with nomic-text-v1.5

The [FineWeb-edu 10BT sample](https://huggingface.co/datasets/HuggingFaceFW/fineweb-edu/tree/main/sample/10BT) was first chunked into 500 tokens (using bert-base-uncased) with 10% overlap resulting in 25 million rows and 10.5BT. 
The chunks were then embedded using [nomic-text-v1.5](https://huggingface.co/nomic-ai/nomic-embed-text-v1.5).

## Dataset Details

### Dataset Description

- **Curated by:** Ian @enjalot Johnson
- **Funded by:** Latent Interfaces
- **License:** Apache license 2.0

### Dataset Sources

- **Repository:** https://github.com/enjalot/fineweb-modal

## Uses

### Direct Use

The dataset was embedded with the `clustering: ` prefix, so the main usecase is clustering and feature extraction. 
The motivation for making the dataset is to create training data for an [SAE to identify features](https://transformer-circuits.pub/2024/scaling-monosemanticity) in nomic-text-v1.5.

## Dataset Structure

The columns of the dataset are:

- id: the document id in fineweb-edu
- url: the url of the document in fineweb-edu
- score: the score from fineweb-edu
- dump: the dump in fineweb-edu
- chunk_index: which chunk of the original document this is
- chunk_text: the text of the chunk
- chunk_tokens: the tokens tokenized by bert-base-uncased
- chunk_token_count: the number of tokens in this chunk
- embedding: the 768 dimension vector representing the nomic-text-v1.5 embedding
## Dataset Creation

### Curation Rationale
The 10BT Sample is big enough to warrant a scaled up process but manageable enough to be done on a small budget. Using on-demand CPUs and GPUs from modal.com the total cost was ~$60.