TL;DR

Transformers.js now supports tool calling with the TextGenerationPipeline, enabling models to interact with external tools, and includes support for the Gemma 4 model (large language model).

New

• Added tool calling support to TextGenerationPipeline.
• Support for Gemma 4 models has been added.

Fixes Worth Knowing

• Generation performance is improved with enhancements to the key-value cache (stores previous calculations).
• Tokenizer (text processing) types are now more accurately determined.

Before You Upgrade

• If you are using pipelines, review the updated default generation parameters to ensure they align with your application's needs.

TL;DR

This release adds support for the Ministral 3 large language model, expanding the range of models available in Transformers.js.

New

• Support for Ministral 3 models.
• Updated Supertonic TTS (text-to-speech) documentation.

Fixes Worth Knowing

• Corrected model naming for Ernie 4.5.

TL;DR

Transformers.js now supports the MobileViTv2 and FastViT models (image classification models), expanding the range of available pre-trained models for image processing tasks.

New

• MobileViTv2 Support: Added support for the MobileViTv2 image classification model.
• FastViT Support: Added support for the FastViT image classification model.

Fixes Worth Knowing

• Images are now automatically rotated for improved processing.
• The library can now read image data directly from blob URIs (data addresses).
• Mistral tokenizer now uses the ungated version for better performance.

TL;DR

Transformers.js has been completely rewritten in ES6, offering improved performance and a simplified API for loading models directly from model IDs (Hugging Face model names).

Breaking

• Model loading now accepts model IDs directly (Hugging Face names) instead of file paths.
• Testing framework updated to Jest (JavaScript testing framework).

New

• Simplified model loading: Use model IDs with pipeline, AutoModel, AutoTokenizer, and AutoProcessor.
• Node.js model caching improves performance for repeated use.

Fixes Worth Knowing

• Improved error handling, including CORS errors.
• Whisper models now calculate mel filters automatically.

Before You Upgrade

• Update your model loading code to use model IDs instead of file paths.
• Review and update any existing tests to align with the new Jest framework.