-
Notifications
You must be signed in to change notification settings - Fork 241
Pull requests: NVIDIA/TransformerEngine
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
[PyTorch] Store FusedAttention's extra_state to DotProductAttention's extra_state when checkpointing
#917
opened Jun 12, 2024 by
cyanguwa
Loading…
3 tasks
Add the option to use SM for P2P comm in TP overlap
#914
opened Jun 11, 2024 by
erhoo82
Loading…
1 of 11 tasks
Fix local cpp tests after inplace build
bug
Something isn't working
#911
opened Jun 11, 2024 by
ksivaman
Loading…
8 of 11 tasks
disable using nvfuser when pytorch version >= 2.2
#905
opened Jun 11, 2024 by
sudhakarsingh27
Loading…
1 of 4 tasks
[Common] Added JIT-compiled fused cast transpose kernels
enhancement
New feature or request
#903
opened Jun 10, 2024 by
Oleg-Goncharov
Loading…
6 of 11 tasks
[C/PyTorch] Removed MPI dependence in Userbuffers
#901
opened Jun 10, 2024 by
denera
Loading…
8 of 11 tasks
[JAX] Splitting cpp_extensions.py
enhancement
New feature or request
jax
#899
opened Jun 7, 2024 by
phu0ngng
Loading…
5 of 11 tasks
[PyTorch] Refine definition of sliding window size based on attention mask
#895
opened Jun 7, 2024 by
cyanguwa
Loading…
3 tasks
Add documentation for dot product attention
#889
opened Jun 4, 2024 by
cyanguwa
Loading…
2 of 4 tasks
Use unoptimized RMSNorm kernel if pointers are not aligned
bug
Something isn't working
#886
opened Jun 3, 2024 by
timmoon10
Loading…
4 of 11 tasks
[Common/PyTorch] Grouped GEMM via multi-stream cuBLAS
#853
opened May 17, 2024 by
yaox12
Loading…
8 of 11 tasks
[Pytorch] Implement fp32 accumulation for attention with context parallel in both forward and backward pass.
#821
opened Apr 28, 2024 by
Yuxin-CV
Loading…
[PyTorch] Fix minor bug in computing num_gqa_groups_per_partition
bug
Something isn't working
#777
opened Apr 13, 2024 by
knowlsie
Loading…
[C/PyTorch] Refactor and move userbuffers into TE/common
#760
opened Apr 8, 2024 by
denera
Loading…
10 of 13 tasks
Previous Next
ProTip!
Filter pull requests by the default branch with base:main.