mn-zerogpu / src

Commit History

  1. fused stacked-expert update kernel + bf16 grad input

    Verified signature
    ede2de3
  2. bf16 stacked-MoE path + BF16/TF32 toggles in prof/glm arms

    Verified signature
    fdc9df4
  3. add prof: arm — torch.profiler + phase split for the 2B MN-GLM step

    Verified signature
    90e87ac

  1. MoE loop-free grad_w (pad+bmm)

    Verified signature
    a82b780
  2. glm arm: swiglu toggle

    Verified signature
    246d3e3
  3. glm arm: reversible toggle + peak-memory reporting

    Verified signature
    d9a8f1f

  1. add glm: MN-GLM (GLM-5.2-class) witness arm

    Verified signature
    ec0eb2c
  2. stacked grouped MoE (batched counter update)

    Verified signature
    37ec00d
  3. grouped-GEMM MoE experts + intg moeNg arm

    Verified signature
    6f3cb01
  4. MoE packed experts (fused update)

    Verified signature
    91f8518
  5. add intg: task — end-to-end integrated GPT(ffn=...) + synced package

    Verified signature
    06d2e6b

  1. Upload folder using huggingface_hub

    Verified signature
    2b9a4ce
  2. Upload folder using huggingface_hub

    Verified signature
    57a73e4
  3. Upload folder using huggingface_hub

    Verified signature
    6f75f79