soulcraft/brainy
1
0
Fork 0
Code Issues 1 Pull requests Projects Releases 335 Packages Wiki Activity Actions

335 releases 522 tags

RSS feed
  • v2.9.0 4d60384755

    v2.9.0 - Clearer Model Precision API Stable

    dpsifr released this 2025-08-29 22:22:28 +02:00 | 990 commits to main since this release

    What's New in v2.9.0

    ✨ Features

    • New precision parameter - Replaces confusing dtype for model variant selection
    • Q8 Model Support Fixed - Properly loads quantized models (75% smaller) when requested
    • Cleaner API - Use precision: 'q8' or precision: 'fp32' for clear model selection

    🔧 Improvements

    • Better model file detection for q8 vs fp32 variants
    • Fixed transformers.js pipeline to correctly load quantized models
    • Updated all internal references to use the new parameter

    📦 Usage

    const brain = new BrainyData({
  embeddingFunction: createEmbeddingFunction({
    precision: 'q8'  // Use 75% smaller quantized models
  })
})

    🚀 Performance

    • Q8 models are 75% smaller (22MB vs 87MB)
    • Maintain 99% accuracy compared to fp32
    • Faster loading and lower memory usage
    Downloads