-
Notifications
You must be signed in to change notification settings - Fork 332
Insights: NVIDIA/TransformerEngine
Overview
Could not load contribution data
Please try again later
3 Pull requests merged by 3 people
-
[JAX] Scale sequence length in CP tests to avoid tiny sizes.
#1347 merged
Dec 4, 2024 -
Improving communication overlap for the case of multi kernel queue usage
#1308 merged
Dec 2, 2024 -
Update list of CI users
#1340 merged
Dec 2, 2024
6 Pull requests opened by 6 people
-
[JAX] Fused attention unit tests fixes and refinements
#1352 opened
Dec 2, 2024 -
Fix attention mask type for Flash Attention + CP + THD
#1354 opened
Dec 4, 2024 -
Add paged attention support
#1355 opened
Dec 4, 2024 -
[JAX] Move parallel encoder tests to L0 distributed test set.
#1356 opened
Dec 4, 2024 -
Disable FP8 in Mcore integration test on older GPUs
#1357 opened
Dec 5, 2024 -
Enabling FP8 all-gather for TE Float8Tensor when using Torch FSDP2
#1358 opened
Dec 5, 2024
1 Issue closed by 1 person
-
Can this project support jetson orin nx?
#1351 closed
Dec 2, 2024
3 Issues opened by 3 people
-
overlapping issue about backward of LayerNormLinear
#1353 opened
Dec 3, 2024 -
te.TransformerLayer fails on H100 with cudnn errors.
#1350 opened
Nov 30, 2024 -
Support more than 1 shape/attention_params for DotProductAttention decision cache
#1349 opened
Nov 29, 2024
6 Unresolved conversations
Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.
-
[C] Normalization Refactor + Adding CUDNN backend
#1315 commented on
Dec 5, 2024 • 12 new comments -
[PyTorch] Bugfix for wgrad bulk overlap conflict when dgrad overlap is reduce-scatter
#1341 commented on
Dec 3, 2024 • 5 new comments -
Bulid transformer enginer is failed caused by cmake command error!
#1020 commented on
Dec 3, 2024 • 0 new comments -
fused out correction in CP
#1248 commented on
Dec 1, 2024 • 0 new comments -
[JAX] Collective GEMM custom op with `nvte_cublas_gemm` (no comm. overlap)
#1307 commented on
Dec 3, 2024 • 0 new comments -
[C/JAX] Comm+GEMM Overlap API for TE/JAX
#1337 commented on
Dec 3, 2024 • 0 new comments