| ---
|
| license: apache-2.0
|
| tags:
|
| - adamw
|
| - base
|
| - classification
|
| - constant-warmup
|
| - perceiver
|
| - relu
|
| - scalenorm
|
| - standard
|
| - trunc-normal
|
| - tucker
|
| ---
|
|
|
| # predict.py
|
|
|
| ## Model Overview
|
|
|
| A **base**-scale implementation of the **perceiver** architecture, built for **classification** tasks.
|
|
|
| ## Architecture
|
|
|
| - **Architecture**: perceiver
|
| - **Scale**: base
|
| - **Attention**: standard
|
| - **Fusion strategy**: tucker
|
| - **Task head**: classification
|
| - **Activation**: relu
|
| - **Normalization**: scalenorm
|
| - **Initialization**: trunc normal
|
|
|
| ## Training
|
|
|
| - **Optimizer**: adamw
|
| - **LR scheduler**: constant warmup
|
|
|
| ## Files
|
|
|
| - `predict.py` — main artifact of this repository
|
|
|
| ## License
|
|
|
| See the license field above.
|
|
|
