metadata
license: mit
tags:
- adafactor
- blip
- contrastive
- exponential
- gated-fusion
- huge
- linear
- rmsnorm
- swish
- xavier-uniform
predict.py
Model Overview
A huge-scale implementation of the blip architecture, built for contrastive tasks.
Architecture
- Architecture: blip
- Scale: huge
- Attention: linear
- Fusion strategy: gated fusion
- Task head: contrastive
- Activation: swish
- Normalization: rmsnorm
- Initialization: xavier uniform
Training
- Optimizer: adafactor
- LR scheduler: exponential
Files
predict.py— main artifact of this repository
License
See the license field above.