Commit History

  1. Fix persist_layer_norm override: set args.no_persist_layer_norm=True (Megatron uses the negated form) Running

    multimodalart HF Staff
    Verified signature
    da5ce3b
  2. Pre-install pybind11 via current interpreter + pre-compile Megatron dataset helpers at module load

    multimodalart HF Staff
    Verified signature
    e39a2c5
  3. Revert UI to plain Gradio; add args.persist_layer_norm=False (torch_norm asserts it false)

    multimodalart HF Staff
    Verified signature
    bbc4bbe
  4. Override TE/Apex fusion flags after initialize_megatron — switch transformer_impl to local, disable all fusion paths

    multimodalart HF Staff
    Verified signature
    0a0f4b8
  5. Add pybind11 (Megatron datasets builder needs python3 -m pybind11 --includes)

    multimodalart HF Staff
    Verified signature
    df280ef
  6. Initial best-effort ZeroGPU port of Khala song generation

    multimodalart HF Staff
    Verified signature
    d1f1097
  7. initial commit

    multimodalart HF Staff
    Verified signature
    0136aab