metadata
license: apache-2.0
tags:
- base
- bilinear
- coca
- constant-warmup
- groupnorm
- multi-query
- multitask
- novograd
- swish
- xavier
model_277377959_coca_base.py
Model Overview
A base-scale implementation of the coca architecture, built for multitask tasks.
Architecture
- Architecture: coca
- Scale: base
- Attention: multi query
- Fusion strategy: bilinear
- Task head: multitask
- Activation: swish
- Normalization: groupnorm
- Initialization: xavier
Training
- Optimizer: novograd
- LR scheduler: constant warmup
Files
model_277377959_coca_base.py— main artifact of this repository
License
See the license field above.