Skip to content

v0.0.20230421

Latest
Latest
Compare
Choose a tag to compare
@zetavg zetavg released this 21 Apr 21:18
· 44 commits to main since this release
v0.0.20230421
d2eef14
This commit was signed with the committer’s verified signature. The key has expired.
zetavg Pokai Chang
GPG key ID: 3177B0C43EB793B0
Expired
Learn about vigilant mode.

New Features

Fine-tuning

  • "Continue from model" now supports using a remote model from Hugging Face Hub.
  • New advanced fine-tune arg: lora_modules_to_save
  • New advanced fine-tune arg: 8bit, FP16, BF16, gradient_checkpointing
  • Support passing additional LoraConfig and TrainingArguments.
  • "Load training parameters from selected model" will now also load the prompt template and dataset.
  • Model trainalbe% will be logged to wandb under config model.trainable%, model.all_params and model.trainable_params.
Assets 2