metadata
license: bsd-3-clause
tags:
- adafactor
- base
- exponential
- gelu-tanh
- kaiming
- low-rank
- matching
- mocov3
- scalenorm
- sparse
model_721644311_mocov3_base.py
Model Overview
A base-scale implementation of the mocov3 architecture, built for matching tasks.
Architecture
- Architecture: mocov3
- Scale: base
- Attention: sparse
- Fusion strategy: low rank
- Task head: matching
- Activation: gelu tanh
- Normalization: scalenorm
- Initialization: kaiming
Training
- Optimizer: adafactor
- LR scheduler: exponential
Files
model_721644311_mocov3_base.py— main artifact of this repository
License
See the license field above.