t0-alpha-onnx-fp16 / webgpu-options.js

Commit History

12 days ago

  1. Fix float32 precision issue in input normalization; re-export and revalidate⁠

    Verified signature
    812d2f2

  1. Use FP16 weight storage with FP32 compute for CPU and WebGPU⁠

    Verified signature
    9f5a8a9