---
license: cc-by-4.0
task_categories:
- tabular-regression
tags:
- exafs
- xafs
- materials-science
- spectroscopy
- materials-project
- feff
pretty_name: LLM4EXAFS Labeled Training Data
private: true
---

# LLM4EXAFS Labeled Training Data

Private release candidate for the labeled training dataset used in the manuscript
*Learning Local Structural Parameters from EXAFS Spectra Using a Pretrained Transformer*.

The uploaded file is a gzip-compressed JSONL file. Each line is one spectrum record with:

- `spectrum_id`
- `absorbing_index`
- `x`
- `y`
- `bond_length`
- `coordination_number`

The dataset is a derived labeled training file, not the original raw structural XAS export.
The current code can read this file directly by passing
`--data_path data/xas_labeled.jsonl.gz`.

## Files

- `xas_labeled.jsonl.gz`: compressed labeled training data.
- `checksums.json`: record count, file size, and SHA-256 digest.

## Provenance

The records are derived from computed XAS/EXAFS data associated with Materials Project
entries and FEFF-calculated spectra, then converted into the training format used by this
repository. Materials Project data are distributed under CC BY 4.0 and require appropriate
attribution.

This is a private draft upload. Public release metadata, citation, and DOI should be
updated after the manuscript data and claims are frozen.