Adapter
An adapter is a small trainable module added to a pretrained model to modify its behavior while leaving most of the model’s original parameters unchanged.
LLM Customization and Fine-Tuning
A small trainable component added to a model to modify its behavior without updating all its original parameters.
- How it is used
- Uses adapters to specialize a frozen base model when prompting falls short and full fine-tuning is too costly.
- How deep it goes
- Explains low-rank matrices and frozen weights, then walks through configuration, training, inference, and multi-adapter routing.
- Examples
- Trains a Qwen3-4B IT support adapter and compares base and adapted responses to a photosynthesis prompt.
- What you can do after
- Apply it via PEFT and TRL to train, save, and load a LoRA adapter for inference.
Read how LLM Customization and Fine-Tuning explains adapter →
Building Specialized AI Systems
An adapter is a small trainable neural-network module inserted into a pretrained model to specialize it while leaving most original parameters unchanged.
- How it is used
- Specializes a shared foundation model for domains such as finance and customer support without full-model retraining.
- How deep it goes
- Explains configuration choices and low-rank updates, including optional merging at inference, without deriving the training algorithm.
- Examples
- Demonstrates finance and casual adapters on T5, credit-risk classification on Llama, and sentiment classification with Rotten Tomatoes reviews.
- What you can do after
- Apply it via PEFT libraries to configure, train, save, load, and swap LoRA adapters on pretrained models.
Transfer Learning for Natural Language Processing
An adapter is a small trainable neural-network module inserted between layers of a pretrained network while its original weights remain fixed.
- How it is used
- Uses adapters to adapt pretrained language models to new tasks without updating their original weights.
- How deep it goes
- Explains adapter placement and parameter freezing, then demonstrates loading and inference without detailing adapter internals.
- Examples
- Demonstrates a BERT sentiment adapter on positive and negative sentences.
- What you can do after
- Apply it via AdapterHub to load a pretrained adapter and run sentiment-classification inference.
Read how Transfer Learning for Natural Language Processing explains adapter →
Rearchitecting LLMs
An adapter is a trainable model component that modifies a pretrained model's behavior while leaving most of the original parameters unchanged.
- How it is used
- Reduces fine-tuning cost while adapting a pretrained model and preserving its frozen base weights.
- How deep it goes
- Explains trainable weights, merging, and dynamic loading, then notes switching costs without detailing adapter internals.
- Examples
- Uses saved adapter directories, merged models, and multiple specializations routed over one base model.
- What you can do after
- Apply via Hugging Face PEFT APIs by saving, merging, or dynamically loading adapters.
Also covered in
- LLMs in Production
- Retrieval Augmented Generation, The Foundational IdeasIn early access (MEAP)
What to know first
- fine-tuningFine-tuning is the continued training of a pretrained model on targeted data or objectives to adapt its parameters or behavior to a task, domain, or use.
- low-rank adaptationLow-rank adaptation (LoRA) is a parameter-efficient fine-tuning technique that adapts a model by learning low-rank updates to its weights, typically while keeping most original weights fixed.
- foundation modelA foundation model is a broadly trained machine-learning model whose learned capabilities can be reused or adapted for a range of downstream tasks.
- GPU memoryGPU memory is memory available to a graphics processing unit for storing data and state used in graphics or computation.
Related concepts
- low-rank adaptationused by · Low-rank adaptation (LoRA) is a parameter-efficient fine-tuning technique that adapts a model by learning low-rank updates to its weights, typically while keeping most original weights fixed.
- LoRA scaling factorconfigured by · The LoRA scaling factor is a configuration value that controls the strength of low-rank adaptation updates relative to the original model weights.
- adaptation continuumpart of · A decision framing that treats customization methods as a range of choices rather than a binary yes-or-no decision.
- deployment pipelineused by · A deployment pipeline is a repeatable sequence of stages that builds, tests or validates, packages, and delivers software changes toward release or deployment.
- endpointused by · An endpoint is an addressable interface of a service through which clients access a resource or invoke an operation, typically via a network request.
Questions about adapter
What is adapter?
An adapter is a small trainable module added to a pretrained model to modify its behavior while leaving most of the model’s original parameters unchanged.
Which Manning books explain adapter?
LLM Customization and Fine-Tuning, Building Specialized AI Systems, Transfer Learning for Natural Language Processing and Rearchitecting LLMs.
What should I know before learning adapter?
Manning books that teach adapter usually introduce fine-tuning, low-rank adaptation, foundation model and GPU memory first.
What is adapter part of?
Adapter is part of adaptation continuum. A decision framing that treats customization methods as a range of choices rather than a binary yes-or-no decision.
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.



