Commits · b0704f1d4748ccc4c56bdc367c20eae7a480ce3d · OpenDAS / FastMoE

05 May, 2021 1 commit
- New data placement structure, speedup · b0704f1d
  TiagoMAntunes authored May 05, 2021
  
  b0704f1d
03 May, 2021 1 commit
- New simplified column reduction kernel · f165928a
  TiagoMAntunes authored May 04, 2021
  
  f165928a
27 Apr, 2021 1 commit
- Fixed incorrect shared memory type in column reduce kernel · 03b78901
  TiagoMAntunes authored Apr 27, 2021
  
  03b78901
04 Apr, 2021 1 commit
- New compute kernel for column reduction · 3d1987d1
  TiagoMAntunes authored Apr 04, 2021
  
  3d1987d1
28 Mar, 2021 2 commits
- Fixed indentation · 7001bc3e
  TiagoMAntunes authored Mar 29, 2021
  
  7001bc3e
- Added CUDA kernel for column reduction · 01d888bf
  TiagoMAntunes authored Mar 29, 2021
  
  01d888bf
25 Mar, 2021 1 commit
- Bias moved to CUDA in forward. Basic bias setup for backwards (no kernel yet) · c5f73a0f
  TiagoMAntunes authored Mar 26, 2021
  
  c5f73a0f
23 Mar, 2021 1 commit

TiagoMAntunes authored Mar 23, 2021

Bias now being calculated directly in MOELinear layer. Added corresponding CUDA changes. Updated forward and backward functions of MOELinear

303d0e93

04 Mar, 2021 1 commit
- remove dependency on cuda sample repo · 7d695acb
  Rick Ho authored Mar 04, 2021
  
  7d695acb
04 Feb, 2021 1 commit
- fix pytorch header compilation bug · 79ccb7b6
  Rick Ho authored Feb 04, 2021
  
  79ccb7b6
01 Feb, 2021 1 commit
- swap local scatter and gather kernel functions · d2039fc7
  Rick Ho authored Feb 01, 2021
  
  d2039fc7
29 Jan, 2021 1 commit
- scatter gather kernel support for non-equal shapes and fix tests · 72e9bc9e
  Rick Ho authored Jan 29, 2021
  
  72e9bc9e
28 Jan, 2021 2 commits
- support fp16 · 49c97411
  Rick Ho authored Jan 28, 2021
  
  49c97411
- single node use torch cuda expert count · a526f438
  Rick Ho authored Jan 28, 2021
  
  a526f438