---
license: mit
tags:
- co-attention
- flash
- lamb
- matching
- mish
- onecycle
- rmsnorm
- swin-t
- xavier
- xlarge
---

# train.py

## Model Overview

A **xlarge**-scale implementation of the **swin t** architecture, built for **matching** tasks.

## Architecture

- **Architecture**: swin t
- **Scale**: xlarge
- **Attention**: flash
- **Fusion strategy**: co attention
- **Task head**: matching
- **Activation**: mish
- **Normalization**: rmsnorm
- **Initialization**: xavier

## Training

- **Optimizer**: lamb
- **LR scheduler**: onecycle

## Files

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

## License

See the license field above.