Sebulah's picture
Create README.md
25a87b6 verified
|
Raw History Blame Contribute Delete
801 Bytes
metadata
language:
  - en
task_categories:
  - text-generation
tags:
  - document-layout-analysis
  - xml-generation
  - causal-lm
  - lora
pretty_name: SPRINT DocBank Subset

SPRINT Pre-processed DocBank Subset

This repository contains the pre-processed, tokenized data subsets used to train the SPRINT layout-aware XML generation model.

  • Original Dataset: DocBank by Minghao Li et al.
  • Modifications: Extracted raw text, aligned with XML bounding box annotations, and formatted into instruction-tuned JSONL triplets for LoRA causal language model fine-tuning.
  • Usage: Created for the SPRINT Dissertation Project. All primary credit for the raw data curation belongs to the original DocBank authors.