Commit History

  1. Fix: unbuffered stdout compat with Python 3.11

    Verified signature
    6a5efbf
  2. Fix: force unbuffered stdout for live HF job logs

    Verified signature
    78128f3
  3. Add 26B-A4B MoE config (hidden=2816, head_dim=256, 16 heads)

    Verified signature
    d93d8a9
  4. Fix: add accelerate dep, dtype instead of torch_dtype

    Verified signature
    6e231be
  5. Fix: total_mem -> total_memory

    Verified signature
    a5ca116
  6. Add self-distill + train job script

    Verified signature
    21873b3
  7. Self-distill + train script v2

    Verified signature
    b2a103b
  8. FastMTP E2B training script

    Verified signature
    043a6bd
  9. initial commit

    Verified signature
    2ee363e