---
license: bsd-3-clause
tags:
- cosine
- gelu
- grouped-query
- lion
- low-rank
- mae
- multitask
- nano
- rmsnorm
- xavier
---

# eval.py

## Model Overview

A **nano**-scale implementation of the **mae** architecture, built for **multitask** tasks.

## Architecture

- **Architecture**: mae
- **Scale**: nano
- **Attention**: grouped query
- **Fusion strategy**: low rank
- **Task head**: multitask
- **Activation**: gelu
- **Normalization**: rmsnorm
- **Initialization**: xavier

## Training

- **Optimizer**: lion
- **LR scheduler**: cosine

## Files

- `eval.py` — main artifact of this repository

## License

See the license field above.