metadata
license: mit
tags:
- approx-gelu
- classification
- dino
- layernorm
- low-rank
- rmsprop
- sparse
- step
- trunc-normal
- xlarge
pipeline.py
Model Overview
A xlarge-scale implementation of the dino architecture, built for classification tasks.
Architecture
- Architecture: dino
- Scale: xlarge
- Attention: sparse
- Fusion strategy: low rank
- Task head: classification
- Activation: approx gelu
- Normalization: layernorm
- Initialization: trunc normal
Training
- Optimizer: rmsprop
- LR scheduler: step
Files
pipeline.py— main artifact of this repository
License
See the license field above.