PEFT

A bucket of fine-tuningⓘfine-tuningtrainingContinued training of a pretrained base model on a smaller, task-specific dataset to specialize its behavior without retraining from scratch. Chat about this → Open full entry → techniques that leave the base model frozen and train a small set of new parameters. LoRA is the dominant one; others include adapter layers (Houlsby 2019), prefix tuning, prompt tuning, and IA3. All trade some expressiveness for orders-of-magnitude reductions in trainable parameter count and adapter storage.

The practical benefits compound at deployment. A single base model can serve many fine-tuned personalities or domains by swapping adapters at inferenceⓘinferenceruntimeRunning a trained model to produce outputs (tokens, images, embeddings) from inputs at serving time, as distinct from the gradient updates of training. Chat about this → Open full entry → time, and adapters themselves are megabytes rather than gigabytes. vLLMⓘvLLMruntimeAn open-source inference engine introduced by UC Berkeley in 2023, built around PagedAttention to manage KV cache memory and serve tokens efficiently under load. Chat about this → Open full entry →, SGLangⓘSGLangruntimeAn open inference engine from the LMSYS team featuring RadixAttention for prefix sharing and a structured-generation frontend, particularly strong on agent and tool-calling workloads. Chat about this → Open full entry →, and other open runtimes support per-request adapter selection for multi-tenant LoRAⓘLoRAtrainingA parameter-efficient fine-tuning method that injects small low-rank adapter matrices into a frozen base model, training a tiny fraction of weights instead of the full model. Chat about this → Open full entry → serving.

The Hugging FaceⓘHugging FacetrainingThe model hub, dataset hub, and open-source library suite (Transformers, Datasets, Tokenizers, Accelerate, PEFT, TRL) that anchors the open-AI ecosystem's distribution and tooling layer. Chat about this → Open full entry → PEFT library is the canonical implementation. It maps the same PeftModel interface over LoRAⓘLoRAtrainingA parameter-efficient fine-tuning method that injects small low-rank adapter matrices into a frozen base model, training a tiny fraction of weights instead of the full model. Chat about this → Open full entry →, AdaLoRA, prefix tuning, prompt tuning, IA3, and the related variants, so training code can change methods with a one-line config edit.