Commit History

  1. Try causal-conv1d 1.6.2 post1

    Verified signature
    970532d
  2. Restore no-CCE FLA wrapper with transformers pin

    Verified signature
    ffe6a1e
  3. Pin transformers and remove CCE

    Verified signature
    72f8c36
  4. Pass FLA grouped constexpr args

    Verified signature
    aa49223
  5. Fix FLA grouped value-head backward patch

    Verified signature
    1038952
  6. Make FLA backward patch contiguous

    Verified signature
    492d6f1
  7. Install cut-cross-entropy and log CUDA memory

    Verified signature
    9ff4e9a

  1. enable core scattermoe tilelang dflash fixes

    Verified signature
    715a73d
  2. add diagnostic tensor prefix breakdown

    Verified signature
    a10bbc7
  3. expose model diagnostics in gradio api

    Verified signature
    efc4392
  4. add model diagnostics endpoint

    Verified signature
    065340b

  1. Deploy TorchAO FP8 ZeroGPU prototype

    Verified signature
    fa47cb8

  1. Install causal conv wheel and skip zero-advantage GRPO groups

    Verified signature
    a6ad39a
  2. Use torch cu130 and batch GRPO rollouts

    Verified signature
    5099b9b
  3. Disable DeepGEMM on ZeroGPU FP8 path

    Verified signature
    cf44aec
  4. Move FP8 model to CUDA inside ZeroGPU hot path

    Verified signature
    6b1b6a1
  5. Install Qwen3.6 FP8 fast path dependencies

    Verified signature
    0b30dcb