v0.2.0

Whisper large tuning using PEFT LoRA+INT-8 on T4 GPU in Colab notebooks

We tested PEFT on @OpenAI's Whisper Large model and got: i) 5x larger batch sizes ii) Less than 8GB GPU VRAM iii) Best part? Almost no degredation to WER 🤯

Without PEFT:

• OOM on a T4 GPU ❌
• 6GB checkpoint ❌
• 13.64 WER ✅

With PEFT:

• Train on a T4 GPU ✅
• 60MB checkpoint ✅
• 14.01 WER ✅

• adding whisper large peft+int8 training example by @pacman100 in https://github.com/huggingface/peft/pull/95

prepare_for_int8_training utility

This utility enables preprocessing the base model to be ready for INT8 training.

• [core] add prepare_model_for_training by @younesbelkada in https://github.com/huggingface/peft/pull/85
• [core] Some changes with prepare_model_for_training & few fixes by @younesbelkada in https://github.com/huggingface/peft/pull/105

disable_adapter() context manager

Enables to disable adapter layers to get the outputs from the frozen base models. An exciting application of this feature allows only a single model copy to be used for policy model and reference model generations in RLHF.

• add disable adapter context manager by @pacman100 in https://github.com/huggingface/peft/pull/106

What's Changed

• release v0.2.0.dev0 by @pacman100 in https://github.com/huggingface/peft/pull/69
• Update README.md by @sayakpaul in https://github.com/huggingface/peft/pull/72
• Fixed typo in Readme by @Muhtasham in https://github.com/huggingface/peft/pull/73
• Update README.md by @pacman100 in https://github.com/huggingface/peft/pull/77
• convert prompt tuning vocab to fp32 by @mayank31398 in https://github.com/huggingface/peft/pull/68
• [core] add prepare_model_for_training by @younesbelkada in https://github.com/huggingface/peft/pull/85
• [bnb] add flan-t5 example by @younesbelkada in https://github.com/huggingface/peft/pull/86
• making prepare_model_for_training flexible by @pacman100 in https://github.com/huggingface/peft/pull/90
• adding whisper large peft+int8 training example by @pacman100 in https://github.com/huggingface/peft/pull/95
• making bnb optional by @pacman100 in https://github.com/huggingface/peft/pull/97
• add support for regex target modules in lora by @pacman100 in https://github.com/huggingface/peft/pull/104
• [core] Some changes with prepare_model_for_training & few fixes by @younesbelkada in https://github.com/huggingface/peft/pull/105
• Fix typo by @mrm8488 in https://github.com/huggingface/peft/pull/107
• add disable adapter context manager by @pacman100 in https://github.com/huggingface/peft/pull/106
• add EleutherAI/gpt-neox-20b to support matrix by @pacman100 in https://github.com/huggingface/peft/pull/109
• fix merging lora weights for inference by @pacman100 in https://github.com/huggingface/peft/pull/117
• [core] Fix autocast issue by @younesbelkada in https://github.com/huggingface/peft/pull/121
• fixes prepare_for_int8_training by @pacman100 in https://github.com/huggingface/peft/pull/127
• issue#126: torch.load device issue. by @gabinguo in https://github.com/huggingface/peft/pull/134
• fix: count params when zero init'd by @zanussbaum in https://github.com/huggingface/peft/pull/140
• chore: update pyproject.toml by @SauravMaheshkar in https://github.com/huggingface/peft/pull/125
• support option for encoder only prompts by @mayank31398 in https://github.com/huggingface/peft/pull/150
• minor fixes to the examples by @pacman100 in https://github.com/huggingface/peft/pull/149
• Add local saving for whisper largev2 example notebook by @alvanli in https://github.com/huggingface/peft/pull/163
• fix count by @dumpmemory in https://github.com/huggingface/peft/pull/162
• Add Prefix Tuning citation by @zphang in https://github.com/huggingface/peft/pull/159
• lora fixes and adding 8bitMegredLinear lora by @pacman100 in https://github.com/huggingface/peft/pull/157
• Update README.md by @pacman100 in https://github.com/huggingface/peft/pull/164
• minor changes by @pacman100 in https://github.com/huggingface/peft/pull/165

New Contributors

• @Muhtasham made their first contribution in https://github.com/huggingface/peft/pull/73
• @mayank31398 made their first contribution in https://github.com/huggingface/peft/pull/68
• @mrm8488 made their first contribution in https://github.com/huggingface/peft/pull/107
• @gabinguo made their first contribution in https://github.com/huggingface/peft/pull/134
• @zanussbaum made their first contribution in https://github.com/huggingface/peft/pull/140
• @SauravMaheshkar made their first contribution in https://github.com/huggingface/peft/pull/125
• @alvanli made their first contribution in https://github.com/huggingface/peft/pull/163
• @dumpmemory made their first contribution in https://github.com/huggingface/peft/pull/162
• @zphang made their first contribution in https://github.com/huggingface/peft/pull/159

Significant community contributions

The following contributors have made significant changes to the library over the last release:

• @mayank31398
  • Prompt Tuning method enhancements and fixes (#68, #150)

Full Changelog: https://github.com/huggingface/peft/compare/v0.1.0...v0.2.0