transformers

huggingface/transformers

Hugging Face Standard Library for Pretrained Multimodal Modelsready

Transformers is Hugging Face's core Python library that unifies the definition and usage of text, image, audio, video, and multimodal models. It is used by researchers, ML engineers, and application developers to load pretrained models and integrate them into training, inference, and serving pipelines. Releases of this package directly impact new model support, API compatibility, and integration stability with PyTorch/JAX/TF and various inference and training ecosystems.

machine-learningnlpcomputer-visionmultimodalpretrained-modelshuggingface

Choose version

5.15.11mo ago

huggingface/transformers Patch release: v5.15.1

Original

Key point

This patch release fixes DFlash and MTP candidate generator errors, as well as image processing issues in accelerator environments.

Details

Bug fix: Resolved device mismatch issue for DFlash candidate tokens when using device_map="auto".
Bug fix: Aligned logit distribution for CandidateGenerators using sampling.
Bug fix: Fixed MTP configuration error when mlp_layer_types is missing.
Bug fix: Changed to fall back to bicubic to prevent image processing failures when using the Lanczos filter in CUDA accelerator environments.
Bug fix: Fixed device movement issue for Gemma4 videos.

transformers releases | AI Briefing