- 11 Sep, 2023 4 commits
- 02 Sep, 2023 1 commit
-
-
zms1999 authored
-
- 25 Aug, 2023 2 commits
- 21 Aug, 2023 1 commit
-
-
Rick Ho authored
fix cublas gemm call for bf16 input
-
- 20 Aug, 2023 1 commit
-
-
xptree authored
-
- 30 Jul, 2023 2 commits
-
-
Rick Ho authored
Update outdated README
-
Mingshu Zhai authored
As mentioned in https://github.com/laekov/fastmoe/issues/166
-
- 24 Jul, 2023 4 commits
- 04 Jul, 2023 1 commit
-
-
Jiezhong Qiu authored
* support megatron v3.0.2 * keep num_experts for lower version of megatron
-
- 06 Jun, 2023 2 commits
-
-
Mingshu Zhai authored
bf16 support
-
Rick Ho authored
-
- 30 May, 2023 2 commits
- 29 May, 2023 1 commit
-
-
Rick Ho authored
-
- 21 May, 2023 5 commits
-
-
Rick Ho authored
Added link to installation guide
-
Rick Ho authored
Create installation-guide.md
-
Arturo Ghinassi authored
-
Arturo Ghinassi authored
-
Arturo Ghinassi authored
Added installation tutorial in the documentation
-
- 19 May, 2023 6 commits
-
-
Rick Ho authored
Cast input to weights type for AMP support
-
Arturo Ghinassi authored
-
Rick Ho authored
Revert "convert input to same type as weight for mixed precision training"
-
Rick Ho authored
-
Rick Ho authored
convert input to same type as weight for mixed precision training
-
Arturo Ghinassi authored
AMP support
-
- 18 May, 2023 1 commit
-
-
Arturo Ghinassi authored
When using CUDA AMP FMoE Linear throws type error as input is half() and weights are float()
-
- 21 Mar, 2023 3 commits
- 20 Mar, 2023 1 commit
-
-
zms1999 authored
-
- 13 Feb, 2023 2 commits
-
-
Rick Ho authored
remove synchronize
-
Fragile-azalea authored
-
- 28 Dec, 2022 1 commit
-
-
Rick Ho authored
Compatibility to older cuda and torch 1.13
-
