Is it possible to train setfit models with the Hugging Face Trainer so that we can have control over hyperparameters such as, for example, b16, optimiser (eg, I prefer to use an 8-bit implementation of AdEMAMix), optimiser hyperparameters (eg, Adam beta1, beta2, AdEMAMix T and beta3), data collator, etc...
Is it possible to train
setfit
models with the Hugging Face Trainer so that we can have control over hyperparameters such as, for example, b16, optimiser (eg, I prefer to use an 8-bit implementation of AdEMAMix), optimiser hyperparameters (eg, Adam beta1, beta2, AdEMAMix T and beta3), data collator, etc...