A field guide to fine-tuning without full-model bloat

What it does

This repository is an awesome-list survey that sorts parameter-efficient fine-tuning research into taxonomies—addition-based, specification-based, reparameterization-based, and unified tuning. Each bucket links to relevant papers (adapters, prompt tuning, LoRA, BitFit, and others) with running paper counts. It is a curated reading list, not a framework.

The interesting bit

Rather than dumping every preprint into one flat list, the maintainers group methods by how they intervene on pretrained weights. That mechanical taxonomy turns a firehose of PEFT literature into something you can reason about by architecture.

Key highlights

• Tracks papers across subcategories like Adapter Tuning (30), Prompt Tuning (27), and LoRA variants (26)
• Organized by intervention type rather than application domain
• Starred by 587 users as a community reference
• Tagged for computer-vision, deep-learning, and transformer workflows
• Maintained with visible commit activity and issue tracking

Caveats

• The README leans heavily on decorative fonts and SVG badges, so the signal-to-noise ratio for skimming is lower than a plain wiki
• It is a bibliography, not a library: you will not find installable code or unified benchmarks here

Verdict

Keep this open if you are surveying the PEFT landscape or choosing between adapters and low-rank updates for a project. Look elsewhere if you need a drop-in training framework.