metadata
license: cc-by-4.0
tags:
- blip
- constant-warmup
- contrastive
- novograd
- scalenorm
- small
- sparse
- swish
- tucker
- xavier
model_159405783_blip_small.py
Model Overview
A small-scale implementation of the blip architecture, built for contrastive tasks.
Architecture
- Architecture: blip
- Scale: small
- Attention: sparse
- Fusion strategy: tucker
- Task head: contrastive
- Activation: swish
- Normalization: scalenorm
- Initialization: xavier
Training
- Optimizer: novograd
- LR scheduler: constant warmup
Files
model_159405783_blip_small.py— main artifact of this repository
License
See the license field above.