v0.3.1: Wave2Vec2, ConvNeXT and BART update

New model additions

• The Wave2Vec2 architecture is now supported for pretraining and the CTC task (#81 and #123)
• The ConvNeXT architecture is now supported for pretraining and the image classification task (#113)
• The sequence classification task is supported for BART (#134, #137, #138)

Fixes

• Speed metrics are now properly computed when resuming from checkpoint (#105)
• The number of samples for the speed metrics are now correct (#109)

v0.3.0: Training transformers models on IPUs

This release can be considered the first official optimum-graphcore release.
It provides:

• A IPU specific config class, IPUConfig, enabling the user to both specify various parameters concerning the IPU and share it on the HuggingFace Hub
• A custom trainer class, IPUTrainer, making training on IPUs seemless for the user compared to the regular transformers Trainer
• A set of example scripts for all the supported tasks

Supported Architectures

The following model architectures can be trained and evaluated using the IPUTrainer:

• Bert
• Roberta
• Deberta
• Lxmert
• Hubert
• ViT

Training only

The following model architectures can be trained (and evaluated without generation):

• Bart
• T5
• GPT-2

Training Scripts

• Language modeling
• Text classification
• Token classification
• Multiple choice
• Question answering
• Summarization
• Translation
• Audio classification
• Image classification