composer.models
Models provided to Trainer
must use the basic
interface specified by BaseMosaicModel
.
Additionally, for convience we provide a number of extensions of BaseMosaicModel
as detailed below.
Base Models
The minimal interface needed to use a model with |
|
Implements the base logic that all classifiers can build on top of. |
|
Implements the base logic that all Transformers can build on top of. |
Image Models
A ResNet-56 model extending |
|
A simple convolutional neural network. |
|
An EfficientNet-b0 model extending |
|
A ResNet-18 model extending |
|
A ResNet-50 model extending |
|
A ResNet-101 model extending |
|
A U-Net model extending |
Language Models
Implements a GPT-2 wrapper around a MosaicTransformer. |
Metrics and Loss Functions
Evaluation metrics for common tasks are
in torchmetrics
and are directly compatible with BaseMosaicModel
.
Additionally, we provide implementations of the following metrics and loss functions.
The Dice Coefficient for evaluating image segmentation. |
|
Torchmetric cross entropy loss implementation. |
|
Drop-in replacement for |
|
Hugging Face compatible cross entropy loss. |
|
Subclasses |