Commit History

2 days ago

  1. Add the quantized codec checkpoints as safetensors: decoder int8, encoder int8 and int4

    Verified signature
    58e6867

3 days ago

  1. XNNPACK decoder: channels-last graph, block 4 in fp16, 512 frames (386 MB peak instead of 2.5 GB)

    Verified signature
    81abbdd

9 days ago

  1. Add the ExecuTorch Vulkan fp16 decoder⁠

    IS02 +1: Claude Opus 5.5
    Verified signature
    5ef58b7
  2. Add the XNNPACK decoder: fp16, its last block and output stage in fp32, with get_model_schema

    Verified signature
    8b11807

10 days ago

  1. Add get_model_schema to the programs (dynamic lengths for react-native-executorch)

    Verified signature
    acd1f7f

11 days ago

  1. Codec programs with one method, forward (was decode / encode); outputs identical

    Verified signature
    4e3104a
  2. Revert the plain weights: remove decoder/plain and encoder/plain, README as at 10e8a97

    Verified signature
    71f89db
  3. Add the plain weights: decoder fp16 and int8, encoder int8 and int4; README

    Verified signature
    bd32746
  4. Lay the programs out as encoder/coreml and decoder/coreml

    Verified signature
    10e8a97