---
language:
- en
pretty_name: Raw Keyword Filtered RedPajamas Dataset
task_categories:
- text-generation
tags:
- marketing
size_categories:
- 1B<n<10B
---

### Getting Started
The dataset is built from the redpajamas dataset after filtering by marketing keywords list that can be found [here](https://github.com/marktrix/redpajama-data-filter-script/blob/main/marketing_words.txt)

The full scripts to recreate the raw dataset before sharding can be found [here](https://github.com/marktrix/redpajama-data-filter-script).

The dataset includes:
- ~4.8B tokens from raw contents.


#### Downloading the dataset

To start exploring and get to know the dataset you can run the script:
```python
import datasets

ds = datasets.load_dataset("marketeam/raw_redpajamas", split="train")

for sample in ds:
  print(sample) # to print the first sample
```

alternatively, you can also use streaming:

```python
import datasets

ds = datasets.load_dataset("marketeam/raw_redpajamas", split="train", streaming=True)

for sample in ds:
  print(sample)
  break # to print the first sample
```


#### Languages

Engish


#### Data Structure
```
├── data
    ├── data-0000.json
    ├── ...
    ├── data-0003.json
```


#### Document structure

```json
{
  "url": "...",
  "date_download": "2023-03-20T08:44:39Z",
  "digest": "sha1:EJNCO5XXIZLG2E3BULUGWCLLJUP2AV2Q",
  "length": 6851,
  "nlines": 49,
  "source_domain": "fenndesign.com",
  "title": "...",
  "raw_content": "...",
  "cc_segment": "...",
  "original_nlines": 101,
  "original_length": 8192,
  "line_ids": [
    6,
    9,
    10,
    11
  ],
  "language": "en",
  "language_score": 0.9,
  "perplexity": 303.6,
  "bucket": "head",
  "id": "2023-14/0000/en_head.json.gz/25",
  "id_int": 4918268498184253468,
  "metadata": {
    "cc_segment": "...",
    "cc_net_source": "2023-14/0000/en_head.json.gz",
    "url": "...",
    "source_domain": "fenndesign.com",
    "language": "en",
    "snapshot_id": "2023-14"
  },
  "quality_signals": {
    "ccnet_length": [
      [
        0,
        6851,
        6851.0
      ]
    ],
    "ccnet_original_length": [
      [
        0,
        6851,
        8192.0
      ]
    ],
    "ccnet_nlines": [
      [
        0,
        6851,
        49.0
      ]
    ],
    "ccnet_original_nlines": [
      [
        0,
        6851,
        101.0
      ]
    ],
    "ccnet_language_score": [
      [
        0,
        6851,
        0.9
      ]
    ]
  },
  "is_duplicate": false
}
```

Document quality annotations can be found [here](https://huggingface.co/datasets/togethercomputer/RedPajama-Data-V2#quality-annotations)