| ---
|
| license: cc-by-4.0
|
| tags:
|
| - concat-mlp
|
| - deit
|
| - generation
|
| - huge
|
| - instancenorm
|
| - kaiming-normal
|
| - onecycle
|
| - relu
|
| - sgd
|
| - standard
|
| ---
|
|
|
| # finetune.py
|
|
|
| ## Model Overview
|
|
|
| A **huge**-scale implementation of the **deit** architecture, built for **generation** tasks.
|
|
|
| ## Architecture
|
|
|
| - **Architecture**: deit
|
| - **Scale**: huge
|
| - **Attention**: standard
|
| - **Fusion strategy**: concat mlp
|
| - **Task head**: generation
|
| - **Activation**: relu
|
| - **Normalization**: instancenorm
|
| - **Initialization**: kaiming normal
|
|
|
| ## Training
|
|
|
| - **Optimizer**: sgd
|
| - **LR scheduler**: onecycle
|
|
|
| ## Files
|
|
|
| - `finetune.py` — main artifact of this repository
|
|
|
| ## License
|
|
|
| See the license field above.
|
|
|
