train.py
Model Overview
A xlarge-scale implementation of the beit architecture, built for matching tasks.
Architecture
- Architecture: beit
- Scale: xlarge
- Attention: standard
- Fusion strategy: concat mlp
- Task head: matching
- Activation: approx gelu
- Normalization: groupnorm
- Initialization: kaiming
Training
- Optimizer: rmsprop
- LR scheduler: polynomial
Files
train.py — main artifact of this repository
License
See the license field above.
