Datasets:

Modalities:
Text
Formats:
parquet
Languages:
English
Libraries:
Datasets
pandas
License:
File size: 1,498 Bytes
83e6d22
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a3ce075
 
 
 
 
f976730
 
 
bd9b050
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f976730
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
---
dataset_info:
  features:
  - name: text
    dtype: string
  - name: source
    dtype: string
  - name: input_ids
    sequence: int32
  splits:
  - name: val
    num_bytes: 481445317
    num_examples: 29016
  download_size: 240052852
  dataset_size: 481445317
configs:
- config_name: default
  data_files:
  - split: val
    path: data/val-*
license: apache-2.0
task_categories:
- text2text-generation
language:
- en
pretty_name: Pretokenized Paloma Dataset
size_categories:
- 10M<n<100M
---

# The Pretokenized Paloma Benchmark Dataset

This dataset is a compact, pre-tokenized evaluation dataset designed to complement the [pretokenized-dolma](https://huggingface.co/datasets/pico-lm/pretokenized-dolma) training set. Built from the [Paloma corpus](https://github.com/allenai/OLMo-Eval/blob/main/paloma/README.md) (Allen Institute), this benchmark was designed to not contain any data overlap with Dolma and is ideal for evaluating models trained on it.

### Overview 

Features:
- Pre-tokenized with the same tokenizer as pretokenized-dolma: [allenai/OLMo-7B-0724-hf](https://huggingface.co/allenai/OLMo-7B-0724-hf)
- Sequence length: 2048 tokens
- Ideal for perplexity calculations for models trained on pretokenized-dolma

We release the exact scripts we use to create this dataset in our [pico-lm/pico-dataset](https://github.com/pico-lm/pico-dataset) GitHub repo.

### Usage

```
from datasets import load_dataset
dataset = load_dataset("pico-lm/pretokenized-paloma", streaming=True)
```