Training model on low RAM GPU

Shivanandroy / simpleT5

simpleT5 is built on top of PyTorch-lightning⚡️ and Transformers🤗 that lets you quickly train your T5 models.

MIT License

382 stars 61 forks source link

Training model on low RAM GPU #48

Open mbledkowski opened 1 year ago

mbledkowski commented 1 year ago

Hello, I am new to machine learning. Is there any way to train T5-large, and bigger models on GPUs with limited RAM - for example by using computer RAM? Best regards, Maciej Błędkowski

EdGaere commented 1 year ago

My suggestion would be to limit the batch size to 6 or 12, so that the model and data fit on the GPU