Commit History

  1. Add direct-load BF16 repeat comparison

    Verified signature
    647387b
  2. Add bounded larger-family CUDA quantizer controls

    Verified signature
    bcba9be
  3. Load 8B directly inside allocated ZeroGPU call

    Verified signature
    7740aa9

  1. Switch larger-family control to root-loaded Qwen3 8B

    Verified signature
    bd2dad5
  2. Disable unused KV cache for embedding controls

    Verified signature
    97e5e64
  3. Expose larger-family benchmark traceback

    Verified signature
    2fad4af
  4. Isolate GTE startup packing from completed 0.6B model

    Verified signature
    234931c
  5. Move GTE 1.5B CUDA placement to ZeroGPU startup

    Verified signature
    88534d9
  6. Add mounted 1.5B and 8B BF16 cross-runtime controls

    Verified signature
    7973967
  7. Lazy-load CUDA quantizer configuration inside GPU call

    Verified signature
    22338c1
  8. Add bounded bitsandbytes INT8 and NF4 controls

    Verified signature
    1a6347c
  9. Add bounded ZeroGPU BF16 CUDA control

    Verified signature
    bc3f111