metadata
license: apache-2.0
tags:
- adam
- approx-gelu
- bilinear
- classification
- coca
- huge
- instancenorm
- kaiming-normal
- polynomial
- standard
eval.py
Model Overview
A huge-scale implementation of the coca architecture, built for classification tasks.
Architecture
- Architecture: coca
- Scale: huge
- Attention: standard
- Fusion strategy: bilinear
- Task head: classification
- Activation: approx gelu
- Normalization: instancenorm
- Initialization: kaiming normal
Training
- Optimizer: adam
- LR scheduler: polynomial
Files
eval.py— main artifact of this repository
License
See the license field above.