Adapters v1.0.0

[calpt]

Blog post: https://adapterhub.ml/blog/2024/08/adapters-update-reft-qlora-merging-models

This version is built for Hugging Face Transformers v4.43.x.

New Adapter Methods & Model Support

• Add Representation Fine-Tuning (ReFT) implementation (LoReFT, NoReFT, DiReFT) (@calpt via Add ReFT (LoReFT, NoReFT, DiReFT) #705)
• Add LoRA weight merging with Task Arithmetics (@lenglaender via Add support for Task Arithmetics #698)
• Add Whisper model support + notebook (@TimoImhof via Add support for Whisper #693; @julian-fong via Adding a notebook for adapters whisper support #717)
• Add Mistral model support (@KorventennFR via Add mistral support  #609)
• Add PLBart model support (@FahadEbrahim via PLbart support #709)

Breaking Changes & Deprecations

• Remove support for loading from archived Hub repository (@calpt via Remove loading adapters from 'ah' & source parameter #724)
• Remove deprecated add_fusion() & train_fusion() methods (@calpt via Remove deprecated functionality #714)
• Remove deprecated arguments in push_adapter_to_hub() method (@calpt via Remove loading adapters from 'ah' & source parameter #724)
• Deprecate support for passing Python lists to adapter activation (@calpt via Remove deprecated functionality #714)

Minor Fixes & Changes

• Upgrade supported Transformers version (@calpt & @lenglaender via Upgrade Transformers to v4.41.x #712, Upgrade Transformers to v4.42.x #719, Upgrade Transformers to v4.43.x #727)
• Fix SDPA/ Flash attention support for Llama (@calpt via Fix Llama sdpa/ flash attention + adapters #722)
• Fix gradient checkpointing for Llama and for Bottleneck adapters (@calpt via Don't throw error if ForwardContext is not available #730)

---

This discussion was created from the release Adapters v1.0.0.