---
issue: 1
status: done
tags: model
---
# equivariant-attention

An equivariant transformer with higher-order messages.

## What happened

- 2026-09-07: 0.689 F1 (+0.076 on the baseline)
- 2026-09-15: 0.742 F1 (+0.129 on the baseline)

## Next

Pretrain with denoising.