AI BriefingTransformers.js v4 Released
Key point
Transformers.js v4 introduces a WebGPU backend, enabling high-performance AI model execution in both browser and server environments.
Details
Transformers.js v4 has been officially released via NPM. The biggest change in this version is the introduction of a new WebGPU Runtime, completely rewritten in C++.
The new runtime was optimized through collaboration with the ONNX Runtime team, and supports WebGPU acceleration not only in browsers but also across various JavaScript environments such as Node, Bun, and Deno.
To maximize performance, a new model export strategy was adopted. In particular, by leveraging specialized ONNX Runtime Contrib Operators such as com.microsoft.MultiHeadAttention, the team achieved approximately a 4x speed improvement for BERT-based embedding models.
Supported models and architectures have also been greatly expanded.
- Support for the latest models including GPT-OSS, GraniteMoeHybrid, and Olmo3
- Implementation of advanced architectures such as Mamba (SSM), MLA, and MoE
- Support for running large-scale models exceeding 8B parameters
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.