What is Transformers?
Transformers is an open-source model-definition framework from Hugging Face that provides state-of-the-art machine learning models for text, computer vision, audio, video, and multimodal tasks, covering both inference and training. Its differentiator is that it centralizes model definitions so they are agreed upon across the ecosystem, making any supported model compatible with a wide range of training frameworks, inference engines, and adjacent modeling libraries.
Key Features
- Centralized model definitions: Transformers acts as the pivot across frameworks, so a model definition supported by Transformers is compatible with the majority of training frameworks (Axolotl, Unsloth, DeepSpeed, FSDP, PyTorch-Lightning), inference engines (vLLM, SGLang, TGI), and adjacent modeling libraries (llama.cpp, mlx) that leverage the definition from Transformers.
- Pipeline: A simple and optimized inference class for many machine learning tasks, including text generation, image segmentation, automatic speech recognition, and document question answering.
- Trainer: A comprehensive trainer for PyTorch models that supports mixed precision, torch.compile, and FlashAttention, along with distributed training.
- generate: Fast text generation with large language models (LLMs) and vision language models (VLMs), including support for streaming and multiple decoding strategies.
- Multi-modality coverage: Model definitions span text, computer vision, audio, video, and multimodal models, for both inference and training.
- Large model catalog: Over 1M+ Transformers model checkpoints are available on the Hugging Face Hub, which users can explore to find a model and get started.
- Model Timeline: A resource for discovering the latest text, vision, audio, and multimodal model architectures in Transformers.
- Versioned documentation: Documentation is available across a long list of releases, from v1.0.0 through the current main branch and v5.x series, alongside translations in languages including AR, DE, EN, ES, FR, HI, IT, JA, KO, PT, RO, TR, and ZH.
Use Cases
- ML engineers running inference: A developer needs text generation, image segmentation, automatic speech recognition, or document question answering and uses the Pipeline class for simple, optimized inference across those tasks.
- PyTorch model training: A team training or fine-tuning models uses the Trainer for mixed precision, torch.compile, FlashAttention, and distributed training.
- LLM and VLM application builders: A developer building generative applications uses generate for fast text generation with streaming and multiple decoding strategies.
- Teams standardizing on a model definition: An organization wants its model to work across training frameworks such as DeepSpeed, FSDP, or PyTorch-Lightning and inference engines such as vLLM, SGLang, or TGI, and relies on Transformers as the shared definition.
- Developers exploring new architectures: A practitioner uses the Models Timeline and the Hugging Face Hub to discover the latest text, vision, audio, and multimodal architectures and start from an existing checkpoint.
Who It Is For
Transformers is aimed at machine learning engineers, researchers, and developers who work with pretrained models for inference or training across text, vision, audio, video, and multimodal tasks. It solves the problem of fragmented model definitions by providing a single, simple, customizable, and efficient definition that stays compatible with the broader training and inference ecosystem, so teams do not have to reimplement models for each framework or engine.
Pricing
The source material does not state pricing for Transformers. The documentation describes the project as open source and part of Hugging Face's open source and open science efforts; no paid tiers or trial terms are specified in the provided content.
Frequently Asked Questions
What is Transformers used for?
Transformers is a model-definition framework for state-of-the-art machine learning models in text, computer vision, audio, video, and multimodal tasks, used for both inference and training. It provides classes such as Pipeline for task inference, Trainer for PyTorch training, and generate for text generation with LLMs and VLMs.
Is Transformers free?
The source material does not specify pricing for Transformers. It is presented as part of Hugging Face's open source and open science mission, and the documentation is publicly available, but no explicit free-tier or licensing terms are stated in the provided content.
What frameworks and inference engines work with Transformers?
Because Transformers centralizes model definitions, supported models are compatible with training frameworks such as Axolotl, Unsloth, DeepSpeed, FSDP, and PyTorch-Lightning, inference engines such as vLLM, SGLang, and TGI, and adjacent modeling libraries such as llama.cpp and mlx.
How many models are available for Transformers?
There are over 1M+ Transformers model checkpoints on the Hugging Face Hub. Users can explore the Hub to find a model and use Transformers to get started, and the Models Timeline highlights the latest architectures.
Who should use Transformers?
Transformers suits machine learning engineers, researchers, and developers who need pretrained models for inference or training and want a model definition that stays compatible across training frameworks, inference engines, and modeling libraries. It is particularly relevant for teams working across multiple modalities or moving models between frameworks.