peft

🤗 PEFT (Parameter-Efficient Fine-Tuning) is a library developed by Hugging Face for efficiently adapting large pretrained models to various downstream applications without fine-tuning all of a model's parameters. By only fine-tuning a small number of extra model parameters, PEFT significantly decreases computational and storage costs while yielding performance comparable to fully fine-tuned models. This makes it more accessible to train and store large language models (LLMs) on consumer hardware. The library is integrated with the Transformers, Diffusers, and Accelerate libraries, providing a faster and easier way to load, train, and use large models for inference. PEFT includes quickstart guides, how-to guides for tasks like image classification, causal language modeling, and automatic speech recognition, as well as conceptual guides and technical references.