The backbone of modern AI, re-engineered
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
Transformers acts as the model-definition framework agreed upon across the ecosystem. It centralizes model definitions so they work with multiple training frameworks (Axolotl, Unsloth, DeepSpeed), inference engines (vLLM, SGLang, TGI), and adjacent libraries.
Transformers supports state-of-the-art models across modalities including text generation, image segmentation, automatic speech recognition, document QA, and multimodal tasks. Users can explore the Timeline to discover latest architectures.
Transformers provides pre-trained model checkpoints (1M+ on Hub), unified model definitions across frameworks, and Pipeline API for inference with minimal code. Users can load models and run inference in seconds without building from scratch.
Transformers integrates with PEFT for parameter-efficient fine-tuning, supports mixed precision training, torch.compile optimization, and distributed training via Accelerate. Users can adapt large models on consumer hardware.
Transformers provides pre-trained models that reduce compute cost, time, and environmental impact compared to training from scratch. Each model is reproduced to match original performance and offers state-of-the-art results out-of-the-box.