transformers-collection

• A collection of transformer models built using huggingface for various tasks. Training done using pytorch lightning.
• Datasets, models and tokenizers from hugging face.
• Goal: Get familiar with huggingface and pytorch lightning ecosystems.

Get started

Train Models using the library

• To train models, install using pip: pip install transformers-collection
• check installation: transformers-collection version

Clone project and modify code

To play around with the code clone the repo:

• git clone git@github.com:aadhithya/transformers-collection.git
• Install poetry: pip install poetry
• Intsall dependencies: poetry install

Note: poetry install will create a new venv. Note: poetry/pip install installs CPU version of pytorch if not available, please make sure to install CUDA version if needed.

Train a model

• Create the yaml config file for the model (see configs/sentiment-clf.yml for example).

• train model using: transformers-collection train /path/to/config.yml

• For a list of supported models, see section Supported Models.

Supported Models / Task

The following models are planned:

Task	Model	Default Dataset	Status	Checkpoint
Text Classification	SentimentClassification	emotion	✅	TBD
Text Summarization	-		🗓️ Planned	TBD

ToDo

• Auto push model to huggingface hub with commit ref.
• make models available via transformers pipelines.
• add more models.