Commit History
-
fix: support torch compile via fake tensors
drbhcd5b9c4
-
fix: re upload build
drbhe0fb143 -
fix: bump build
drbh5c51af4 -
fix: add quickstart and avoid autotune when no cuda
drbh09e15a7
-
feat: bump build for shared experts
drbh3bdb4b8 -
feat: support shared experts layer and tests
drbh89e2950
-
fix: extract expert device mesh for group from unused prehook
drbhaa23f77
-
Build from kernel-builder job
Verified signatureeba2c2c -
fix: prefer using passed parallel group
drbh76c7de7 -
feat: full rebuild
drbh9436276 -
fix: adjust layer params in source
drbh9a1816c -
feat: adjust layer params
drbhb08f6c9
-
feat: bump build
drbh5268e56 -
fix: improve expert parallel implementation and refactors
drbhe47036a -
feat: bump build
drbh0e97a7c -
fix: add parallel forward functional logic
drbh13afbbe
-
feat: bump build for fully vendored version
drbh9354548 -
fix: fully vendor stk and fix imports
drbh63599de -
fix: bump build
drbhef966b6 -
fix: vendor stk decorators
drbh0586ba6 -
feat: bump builds
drbh2b84d84 -
feat: add functional version of moe class and layer
drbhdabb815
-
fix: bump build and imports
drbhb2bfc37 -
fix: prefer relative imports
drbh484fde0 -
fix: bump builds
drbh6756875 -
feat: vendor grouped gemm
drbh3224250
-
feat: add build output
drbha585153
-
fix: prefer nix based library inclusion
drbh359242d -
feat: validate build with original test suite
drbh9c4ca75
-
fix: bump flake for github builder
drbha4f6452 -
feat: initial port of megablocks to builder format
drbh2595c46