metadata
license: mit
tags:
- dilated
- gated-fusion
- matching
- mocov3
- polynomial
- rmsnorm
- rmsprop
- swish
- xavier-uniform
- xlarge
run.py
Model Overview
A xlarge-scale implementation of the mocov3 architecture, built for matching tasks.
Architecture
- Architecture: mocov3
- Scale: xlarge
- Attention: dilated
- Fusion strategy: gated fusion
- Task head: matching
- Activation: swish
- Normalization: rmsnorm
- Initialization: xavier uniform
Training
- Optimizer: rmsprop
- LR scheduler: polynomial
Files
run.py— main artifact of this repository
License
See the license field above.