feat: support shared experts layer and tests
89e2950
drbh
commited on
Commit History
fix: extract expert device mesh for group from unused prehook
aa23f77
drbh
commited on
fix: prefer using passed parallel group
76c7de7
drbh
commited on
fix: adjust layer params in source
9a1816c
drbh
commited on
fix: improve expert parallel implementation and refactors
e47036a
drbh
commited on
fix: add parallel forward functional logic
13afbbe
drbh
commited on
fix: fully vendor stk and fix imports
63599de
drbh
commited on
fix: vendor stk decorators
0586ba6
drbh
commited on
feat: add functional version of moe class and layer
dabb815
drbh
commited on
fix: prefer relative imports
484fde0
drbh
commited on
feat: vendor grouped gemm
3224250
drbh
commited on
feat: validate build with original test suite
9c4ca75
drbh
commited on
feat: initial port of megablocks to builder format
2595c46
drbh
commited on