alexandermam's picture
Upload 2 files
0530124 verified
|
Raw
History Blame Contribute Delete
763 Bytes
metadata
license: apache-2.0
tags:
  - batchnorm
  - cross-attention
  - deit
  - generation
  - huge
  - kaiming
  - relu
  - rmsprop
  - sliding-window
  - step

model_587614435_deit_huge.py

Model Overview

A huge-scale implementation of the deit architecture, built for generation tasks.

Architecture

  • Architecture: deit
  • Scale: huge
  • Attention: sliding window
  • Fusion strategy: cross attention
  • Task head: generation
  • Activation: relu
  • Normalization: batchnorm
  • Initialization: kaiming

Training

  • Optimizer: rmsprop
  • LR scheduler: step

Files

  • model_587614435_deit_huge.py — main artifact of this repository

License

See the license field above.