---
dataset_info:
  features:
  - name: input_ids
    list: int32
  - name: length
    dtype: int64
  splits:
  - name: train
    num_bytes: 9591360
    num_examples: 4656
  - name: validation
    num_bytes: 990860
    num_examples: 481
  download_size: 10551849
  dataset_size: 10582220
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
language:
- en
license: cc-by-sa-3.0
task_categories:
- text-generation
tags:
- wikitext
- gpt2
- influence-functions
- data-attribution
size_categories:
- 1K<n<10K
---

# bergson-wikitext-2-4656-chunks

Pre-tokenized WikiText-2 in fixed 512-token GPT-2 chunks, used by the
[bergson](https://github.com/EleutherAI/bergson) WikiText-2 / GPT-2 data-attribution
replication of Bae et al. 2024 ([_Training Data Attribution via Approximate
Unrolled Differentiation_](https://arxiv.org/abs/2405.12186)).

Each row is one 512-token document (one block of the concatenated corpus).

| Split | Rows |
|-------|------|
| `train` | 4656 |
| `validation` | 481 |

Columns:

- `input_ids` — `list[int32]`, length 512 (GPT-2 BPE token ids)
- `length` — `int64`, always 512

Load with `chunk_length: 0` in a bergson config (the data is already tokenized
and chunked).

## How this was produced

The standard HuggingFace `run_clm` preprocessing recipe — the same one
kronfluence's `examples/wikitext` uses — applied to public
[`Salesforce/wikitext`](https://huggingface.co/datasets/Salesforce/wikitext)
`wikitext-2-raw-v1`:

1. Tokenize the raw `text` column with the GPT-2 tokenizer, no added special
   tokens.
2. `group_texts`: within each 1000-row `datasets.map` batch, concatenate the
   tokens, drop the remainder, and slice into fixed 512-token blocks.

The 1000-row batching is load-bearing: it is what yields exactly 4656 / 481
chunks (whole-corpus concatenation instead gives 4671 / 482). No shuffling or
seed is involved — the recipe is deterministic.

```python
from datasets import DatasetDict, load_dataset
from transformers import AutoTokenizer

BLOCK = 512
tok = AutoTokenizer.from_pretrained("gpt2")
raw = load_dataset("Salesforce/wikitext", "wikitext-2-raw-v1")


def group_texts(examples):
    concat = sum(examples["input_ids"], [])
    total = (len(concat) // BLOCK) * BLOCK
    return {"input_ids": [concat[i : i + BLOCK] for i in range(0, total, BLOCK)]}


out = {}
for split in ("train", "validation"):
    toks = raw[split].map(
        lambda b: tok(b["text"]),
        batched=True,
        remove_columns=raw[split].column_names,
    )
    blocks = toks.map(
        group_texts,
        batched=True,
        remove_columns=[c for c in toks.column_names if c != "input_ids"],
    )
    out[split] = blocks.map(lambda x: {"length": len(x["input_ids"])})

DatasetDict(out).push_to_hub("EleutherAI/bergson-wikitext-2-4656-chunks")
```

This is also committed as `examples/replication/prep_dataset.py` in the bergson
repo. The chunking reproduces the reference dataset byte-for-byte (identical
`input_ids` on all 4656 train and 481 validation rows).

## License

Inherits WikiText-2's CC BY-SA 3.0.