Skip to content

Pull requests: ROCm/TransformerEngine

New pull request New
56 Open 605 Closed
56 Open 605 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

gfx942-only build bugfix and kittens refactor ci-level 3 CI test level 3
#698 opened Aug 8, 2026 by alextmagro Contributor Loading…
3
Integrate MXFP4 hipblaslt GEMM support
#697 opened Aug 7, 2026 by VeeraRajasekhar Contributor • Draft
13 tasks
@VeeraRajasekhar
sGPU Test Scheduling: Global Work Queue
#696 opened Aug 7, 2026 by VeeraRajasekhar Contributor • Draft
9 of 13 tasks
@VeeraRajasekhar
Compile CP softmax LSE corrections with dynamic shapes ci-level 1 CI test level 1
#695 opened Aug 6, 2026 by JessicaJiang-123 Loading…
6 of 13 tasks
Permute-Free Grouped GEMM for MoE (bf16, gfx950) ci-level 1 CI test level 1
#694 opened Aug 6, 2026 by sudhu2k Contributor • Draft
8 of 13 tasks
@sudhu2k
IFU 2.17 upstream feature enablement
#692 opened Aug 4, 2026 by AllenFarcas Contributor • Draft
1 of 18 tasks
1
gfx1250: fix MXFP8 scale_inv shape mismatch in C++ tests ci-level 1 CI test level 1
#691 opened Aug 4, 2026 by matthiasdiener Contributor Loading…
13 tasks
@matthiasdiener
Veergopu/upgrade ci rock 714
#690 opened Aug 4, 2026 by VeeraRajasekhar Contributor Loading…
9 of 13 tasks
@VeeraRajasekhar
18
Honor timeout settings in subprocess run wrapper ci-level 2 CI test level 2
#689 opened Aug 1, 2026 by ipanfilo Collaborator Loading…
1 of 13 tasks
2
Consolidating blockwise FP32 scale flag ci-level 3 CI test level 3
#687 opened Jul 31, 2026 by asdfvg123 Contributor Loading…
1 of 13 tasks
6
Updated AITER/QoLA ci-level 3 CI test level 3
#683 opened Jul 28, 2026 by Micky774 Contributor Loading…
13 tasks
3
microbenchmarks: usv implementation
#679 opened Jul 24, 2026 by matthiasdiener Contributor • Draft
13 tasks
[ROCm] Jax Add softmax sink (learnable off-by-one) support for the ROCm/CK fused attention backend ci-level 3 CI test level 3
#678 opened Jul 24, 2026 by shurale-nkn Loading…
2 of 13 tasks
18
microbenchmarks: add buffer rotation option
#677 opened Jul 23, 2026 by matthiasdiener Contributor Loading…
1 of 13 tasks
@matthiasdiener
12
Experimental FlyDSL GEMM backend for TE PyTorch (BF16/FP16/FP32/FP8/MXFP8) ci-level 1 CI test level 1
#676 opened Jul 22, 2026 by aris134 Contributor Loading…
@aris134
20
[wip] native nvfp4 GEMM support on gfx1250 ci-level 1 CI test level 1
#675 opened Jul 20, 2026 by matthiasdiener Contributor • Draft
13 tasks
@matthiasdiener
ci: bump te-rocm-wheels artifact retention 1d -> 5d
#673 opened Jul 17, 2026 by wenchenvincent Collaborator Loading…
2 tasks
3
[proof-of-concept] benchmarks dashboard
#670 opened Jul 14, 2026 by matthiasdiener Contributor • Draft
13 tasks
@matthiasdiener
Experimental Triton GEMM backend for TE PyTorch (BF16/FP16/FP32/FP8/MXFP8) ci-level 3 CI test level 3
#667 opened Jul 9, 2026 by wenchenvincent Collaborator Loading…
4 of 7 tasks
62
Updated CK/AITER Cmake Build ci-level 1 CI test level 1
#666 opened Jul 9, 2026 by Micky774 Contributor Loading…
13 tasks
10
Initial integration of a4w4 GEMM ci-level 3 CI test level 3
#663 opened Jul 7, 2026 by Micky774 Contributor • Draft
13 tasks
8
CI: Fix runners GPU isolation ci-level 3 CI test level 3
#659 opened Jul 7, 2026 by leo-automation Collaborator Loading…
Ipanfilo/aiter split kv fwd ci-level 2 CI test level 2
#655 opened Jun 29, 2026 by ipanfilo Collaborator • Draft
13 tasks
[Feat] Added JAX-Triton bridge for ROCm
#649 opened Jun 24, 2026 by AllenFarcas Contributor Loading…
6 of 13 tasks
@AllenFarcas
13
Relax MXFP8 GEMM K constraint from multiple-of-128 to multiple-of-64
#642 opened Jun 19, 2026 by JohnQinAMD Loading…
13 tasks
1
ProTip! Follow long discussions with comments:>50.