preTrainingNEAR / README.md
jcarbonnell's picture
Upload dataset
b593a1a verified
metadata
language:
  - en
license: apache-2.0
dataset_info:
  features:
    - name: text
      dtype: string
  splits:
    - name: train
      num_bytes: 6855034
      num_examples: 1022
    - name: val
      num_bytes: 850752
      num_examples: 114
  download_size: 3998284
  dataset_size: 7705786
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: val
        path: data/val-*

This dataset is a subset of the original nearData dataset, prepared for the continued pre-training of a pre-trained LLM. The idea behind the continued pre-training of pre-trained models is to further instruct them with specific information, in this case on the Near Protocol blockchain, before fine-tuning them.

The preTrainingNEAR dataset was prepared from local text files using the datasets library from Hugging Face. It includes:

  • nearBlog: 481 blog articles from Near Blog collected on March 13th, 2024.
  • nearBosWebEngine: 13 docs files from the Near BOS Wen Engine collected on May 21st, 2024.
  • nearDocs: 395 docs files from Near Docs collected on March 13th, 2024.
  • nearNEPs: 124 docs files from the NEAR Enhancement Protocol collected on May 21st, 2024.
  • nearNode: 40 docs files from the Near Node Docs collected on May 21st, 2024.
  • nearPapers: 3 papers from the Near Papers collected on May 21st, 2024.
  • nearWiki: 98 docs from the Near Wiki collected on May 21st, 2024.