Snapshot of contributions from the DeepIsaHOL project

Published March 24, 2025 | Version v1

Software Open

This is a (2025-03-24) snapshot of the repository with contributions from the DeepIsaHOL project. This snapshot includes:

1. Sources for retrieving proof data from Isabelle libraries

2. Sources for the read-eval-print-loop interfaces for Isabelle in Scala and Python

3. A Python training loop for Hugging Face T5 models for conditional generation

4. Python sources for evaluating the trained models

5. Examples of the generated data as JSON files

6. Log files of the training and evaluation loops

7. The model weights for all trained models

Files

Name	Size	Download all
itp2025_artefact.zip md5:6b9c5cd509c0c8c33a1b2633872251b1	5.4 GB	Preview Download

European Commission
DeepIsaHOL - Reinforcement learning to improve proof-automation in theorem proving 101102608