README.md

metadata

license: apache-2.0

Interchangeable Token Embeddings - Datasets

This repository contains the datasets for the ICML 2025 paper, Interchangeable Token Embeddings for Extendable Vocabulary and Alpha-Equivalence.

Links:

1. arXiv
2. Project Page
3. Code
4. Models

Structure

• data: Datasets for the LTL task
  • na-*: Datasets from the DeepLTL paper
  • ltl-35-10ap: Dataset for the full vocabulary baseline
  • ltl-35-perturbed: The training dataset used in perturbation experiments
  • eval2d-10ap.pkl: The test dataset used in the heatmap evaluations
• data-prop: Datasets for the propositional logic task
  • Same internal structure

The datasets for the copy task are omitted since those are just random strings and unnecessarily large, but I can upload them upon request.