Kernels
megablocks / torch-ext

Commit History

  1. fix: include torch compile flag

    drbh
    8176cbe
  2. fix: support torch compile via fake tensors

    drbh
    cd5b9c4

  1. fix: add quickstart and avoid autotune when no cuda

    drbh
    09e15a7

  1. feat: support shared experts layer and tests

    drbh
    89e2950

  1. fix: extract expert device mesh for group from unused prehook

    drbh
    aa23f77

  1. fix: prefer using passed parallel group

    drbh
    76c7de7
  2. fix: adjust layer params in source

    drbh
    9a1816c

  1. fix: improve expert parallel implementation and refactors

    drbh
    e47036a
  2. fix: add parallel forward functional logic

    drbh
    13afbbe

  1. fix: fully vendor stk and fix imports

    drbh
    63599de
  2. fix: vendor stk decorators

    drbh
    0586ba6
  3. feat: add functional version of moe class and layer

    drbh
    dabb815

  1. fix: prefer relative imports

    drbh
    484fde0
  2. feat: vendor grouped gemm

    drbh
    3224250

  1. feat: validate build with original test suite

    drbh
    9c4ca75

  1. feat: initial port of megablocks to builder format

    drbh
    2595c46