model_170114501_blip_large.py

Model Overview

A large-scale implementation of the blip architecture, built for multitask tasks.

Architecture

• Architecture: blip
• Scale: large
• Attention: sliding window
• Fusion strategy: low rank
• Task head: multitask
• Activation: gelu tanh
• Normalization: layernorm
• Initialization: xavier

Training

• Optimizer: rmsprop
• LR scheduler: polynomial

Files

• model_170114501_blip_large.py — main artifact of this repository

License

See the license field above.