metadata
license: mit
tags:
- adamw
- batchnorm
- co-attention
- deit
- flash
- gelu
- kaiming-normal
- matching
- nano
- step
inference.py
Model Overview
A nano-scale implementation of the deit architecture, built for matching tasks.
Architecture
- Architecture: deit
- Scale: nano
- Attention: flash
- Fusion strategy: co attention
- Task head: matching
- Activation: gelu
- Normalization: batchnorm
- Initialization: kaiming normal
Training
- Optimizer: adamw
- LR scheduler: step
Files
inference.py— main artifact of this repository
License
See the license field above.