AI Briefing
KO

Transformers.js v3: WebGPU Support

·2024.10.22 09:00

Key point

Transformers.js v3 has been released, maximizing browser-based AI inference performance through WebGPU support.

Details

Hugging Face's Transformers.js v3 has been officially released. The core of this update is WebGPU support, which delivers up to 100x faster performance compared to the existing WASM approach.

The key updates are as follows:

  • WebGPU acceleration: Enables high-performance computation by directly leveraging the GPU in the browser
  • Expanded support: Support for a total of 120 architectures and over 1,200 pre-converted models
  • New quantization formats: Support for various data types (dtypes)
  • Environment compatibility: Full compatibility with Node.js (ESM/CJS), Deno, and Bun
  • Developer convenience: 25 new example projects and templates provided

With just a device: 'webgpu' setting, developers can accelerate various tasks such as text embedding, speech recognition (Whisper), and image classification in the browser environment.

machine-learninghuggingfacejavascripttransformers-jswebgpu
Read originalShare on Threads

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.

Related

HuggingFace BlogHugging Face Releases WebGPU KernelsTLDR AIHugging Face Considers Sale at $13 Billion ValuationPyTorchKRLEADBOARD Benchmark for Drug Prediction ReleasedHuggingFace BlogImplementing a Robot Data Loop Based on Strands and LeRobotPyTorchKRvLLM Releases 'Decision 1.0', AI Models Specialized for Decision-MakingApple MLApple Unveils 'Probe Guidance' Technique to Enhance Flow Matching Model Performance