| | --- |
| | license: mit |
| | task_categories: |
| | - text-generation |
| | language: |
| | - en |
| | size_categories: |
| | - 10M<n<100M |
| | configs: |
| | - config_name: chunk_size_1024 |
| | data_files: "gutenberg_en_paragraph_1024.jsonl" |
| | - config_name: chunk_size_2048 |
| | data_files: "gutenberg_en_paragraph_2048.jsonl" |
| | --- |
| | |
| | **Description of the dataset** |
| |
|
| | This is the November 16, 2023 snapshot of the English subset of the Project Gutenberg corpus (containing 56712 documents in total), downloaded and preprocessed with code from [this repository](https://github.com/eminorhan/gutenberg). |
| |
|
| | Two different versions of the data are provided: |
| |
|
| | * The `chunk_size_1024` version divides the data into ~14.2M records consisting of a few paragraph long chunks of text, where each chunk is at least 1024 chars long, and the corresponding metadata. |
| |
|
| | * The `chunk_size_2048` version divides the data into ~8.2M records consisting of a few paragraph long chunks of text, where each chunk is at least 2048 chars long, and the corresponding metadata. |
| |
|
| | This dataset is ideal for generating fine-grained embeddings of the documents. |
