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

5.16.11mo ago

huggingface/transformers Release v5.16.1

Original

Key point

Includes major fixes such as added support for the GLM-5.3-Flash model and restored Tensor-Parallel API compatibility

Details

Key Features

Added support for GLM-5.3-Flash: the first native multimodal model in the GLM-5 series, adopting a hybrid architecture (combining sparse and linear attention) that activates only 18B of its total 320B parameters, significantly reducing long-context serving costs and boosting efficiency.

Bug Fixes & Compatibility

Restored backward-compatible (BC) behavior of the Tensor-Parallel API.
Fixed kernel commit and repository path errors related to ESMFold2.
Applied version pinning for HF kernel for security purposes.

transformers releases | AI Briefing