Practical overview
When it fits
Use Adapters when developing an NLP workflow with Hugging Face Transformers and PyTorch where you want to train, load, configure or compose adapter methods rather than treat fine-tuning as a single undifferentiated step.
Who it is for
- NLP developers using Hugging Face Transformers and PyTorch
- Researchers exploring parameter-efficient fine-tuning and modular transfer learning
- Teams maintaining model workflows that need to load or compose pretrained adapters
Documented uses
- Add an adapter method to an existing Transformers model and train it.
- Load a pretrained adapter for inference with a model and tokenizer.
- Configure or combine multiple adapters in one model.
- Explore supported adapter techniques, including quantized training options and adapter merging.
Requirements & limits
The repository identifies the project as open source under the Apache-2.0 license. Installation is documented through PyPI or from source.
- The stated minimum requirements are Python 3.9+ and PyTorch 2.0+.
- The README describes integration of 10+ adapter methods and 20+ Transformer models; consult the model and method documentation to check a particular pairing.
- The repository offers both pretrained-adapter loading and training workflows, as well as configuration and composition features.
- The library replaces the older adapter-transformers package; the project provides transition guidance and states that previously trained adapters remain compatible.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Add and train a LoRA adapter on an existing model
Initialize Adapters on an existing Transformers model, add a LoRA adapter, and prepare it for training.
Before you start
- Adapters installed with Python 3.9+ and PyTorch 2.0+.
- A Transformers model that can be loaded as
AutoModelForSequenceClassification.
Steps
Load a model, as in the example:
AutoModelForSequenceClassification.from_pretrained("t5-base").Call
adapters.init(model)to initialize adapter support on the model.Add the adapter with
model.add_adapter("my_lora_adapter", config="lora").Call
model.train_adapter("my_lora_adapter"), then continue with a regular training loop.
Expected outcome
The model has a LoRA adapter configured for adapter training; the repository example leaves the training loop to the user.
Guide limitations
- The source does not provide the details of the regular training loop or report training results.
- Adapter use depends on supported model architectures; consult the model overview for model support.
Compare options for Model inference and serving
These projects document a common task and share a project role. This is a source-based comparison, not an execution benchmark or a guarantee of interchangeability.
| Project | Documented purpose | Platform | Access and requirements |
|---|---|---|---|
| Adapters | Add an adapter method to an existing Transformers model and train it. | The repository identifies the project as open source under the Apache-2.0 license. Installation is documented through PyPI or from source. | |
| Hyperpose | Customize pose-estimation architectures and datasets. | Docker, Linux, macOS, Windows | |
| Multi Task Nlp | Define multiple tasks and their data and model settings in a YAML file. | Linux, macOS, Windows | The project is an open-source library under the Apache-2.0 license, distributed through its public GitHub repository. |
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 14 October 2026
- Original website title
- adapter-hub/adapters · Official homepage title · retained in its source language