TxT360 / README.md
victormiller's picture
Update README.md
83de6dd verified
|
raw
history blame
6.05 kB
metadata
license: odc-by

TxT360: a globally deduplicated dataset for LLM pretraining

k2 eval table

We introduce TxT360 (Trillion eXtracted Text) the first dataset to globally deduplicate 99 CommonCrawl snapshots and 14 commonly used non-web data sources (e.g. FreeLaw, PG-19, etc.) providing pretraining teams with a recipe to easily adjust data weighting and train the most performant models.

TxT360 Compared to Common Pretraining Datasets

Data Source TxT360 FineWeb RefinedWeb PedPajamaV2 C4 Dolma RedPajamaV1 The Pile
CommonCrawl Snapshots 99 96 90 84 1 24 5 0.6% of 74
Papers 5 Sources - - - - 1 Source 1 Source 4 Sources
Wikipedia 310+ Languages - - - - Included Included English Only
FreeLaw Included - - - - - - Included
DM Math Included - - - - - - Included
USPTO Included - - - - - - Included
PG-19 Included - - - - Included Included Included
HackerNews Included - - - - - - Included
Ubuntu IRC Included - - - - - - Included
EuroParl Included - - - - - - Included
StackExchange Included - - - - - - Included
Code ** - - - - Included Included Included

** TxT360 does not include code. This decision was made due to the perceived low duplication code with other sources.

Complete details on the dataset can be found in our blog post here.

Upsampling Experiment with Comparison to FineWeb

To evaluate the training efficiency of our dataset, we sampled 1.5T tokens from both FineWeb and TxT360 (using the aforementioned weighting) and conducted a training ablation on an 8x8B Mixture-of-Experts architecture, similar to Mixtral. We compared the learning curves by tracking training loss, validation scores, and performance across a wide array of diverse evaluation benchmarks. The validation set was sampled independently from SlimPajama. Note that this experiment is done on a slightly earlier version of the dataset.

comparison

Initial Data Representation

To produce TxT360, a comprehensive and transparent data processing pipeline was designed to account for the nuances of both web and curated datasets. The pipeline presents a unified framework for processing both data types, making it convenient and easily adaptive for users to revise and fine-tune the pipeline for their own use cases.

Web datasets are inherently noisy and varied. The TxT360 pipeline implements sophisticated filtering and deduplication techniques to clean and remove redundancies while preserving data integrity.

Curated datasets are typically structured and consistently formatted. TxT360 filters these sources with selective steps to maintain their integrity while providing seamless integration into the larger dataset. Both data source types are globally deduplicated together resulting in 5.7T tokens of high-quality data. The table below shows the source distribution of TxT360 tokens.

Data Source Raw Data Size Token Count Information Cut-Off Date
CommonCrawl 9.2 TB 4.83T 2024-30
Papers 712 GB 154.96B Q4 2023
Wikipedia 210 GB 4.75B -
Freelaw 23 GB 7.34B Q1 2024
DM Math 22 GB 5.23B -
USPTO 45 GB 4.95B Q4 2023
PG-19 11 GB 2.94B -
HackerNews 4.1 GB 1.08B Q4 2023
Ubuntu IRC 4.7 GB 1.54B Q4 2023
Europarl 6.1 GB 1.96B -
StackExchange 45 GB 8.37B Q4 2023

CommonCrawl Data Filtering

Follow this link to view all steps taken to filter the web data.

Curated Source Filtering

Each data source was filtered individually with respect to the underlying data. Full details and discussion on how each source is filter is covered here.

Global Deduplication

After the web and curated sources were filtered, they were globally deduplicated to create TxT360. The deduplication process is available here.

Citation

BibTeX:

@misc{txt360data2024,
      title={TxT360: a globally deduplicated dataset for LLM pretraining}, 
      author={Liping Tang, Nikhil Ranjan, Omkar Pangarkar, Zhen Wang, An Li, Zhoujun Cheng, Suqi Sun, Cun Mu, Victor Miller, Yue Peng, Eric P. Xing, Zhengzhong Liu},
      year={2024}
}