Skip to content

[CI] Enroll the remaining Kernels jobs in automatic sharding (30 jobs) #64591

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
khluu merged 2 commits into from
Oct 30, 2026
Merged

[CI] Enroll the remaining Kernels jobs in automatic sharding (30 jobs) #64591

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
3316362
[CI] Enroll the remaining Kernels jobs in automatic sharding
Thangnguyenvn98 Oct 29, 2026
0d343ff
[CI] Also enroll the MoRI EP Numerics mirror of B200 FusedMoE Layer K…
Thangnguyenvn98 Oct 29, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
32 changes: 31 additions & 1 deletion .buildkite/test_areas/kernels.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,7 @@ depends_on:
steps:
- label: ":nvidia: (H200 MIG 18GB) vLLM IR"
key: vllm-ir-tests
automatic_shard: true
timeout_in_minutes: 35
device: h200_18gb
working_dir: "/vllm-workspace/"
Expand All @@ -15,6 +16,7 @@ steps:
- pytest -v -s tests/kernels/ir
mirror:
amd:
automatic_shard: true
label: ":amd: (MI355 DPX) vLLM IR"
dind: false
device: mi355_dpx
Expand Down Expand Up @@ -83,6 +85,7 @@ steps:

- label: ":nvidia: (H100) MiniMax Reduce RMS Kernels"
key: kernels-minimax-reduce-rms-test-2-gpus
automatic_shard: true
timeout_in_minutes: 20
num_devices: 2
device: h100
Expand All @@ -98,6 +101,7 @@ steps:
- pytest -v -s kernels/core/test_minimax_reduce_rms.py
mirror:
amd:
automatic_shard: true
label: ":amd: (MI355) MiniMax Reduce RMS Kernels"
dind: false
device: mi355_2
Expand Down Expand Up @@ -146,6 +150,7 @@ steps:

- label: ":nvidia: (H100) BF16x3 Router GEMM Kernels"
key: kernels-bf16x3-router-gemm-h100
automatic_shard: true
timeout_in_minutes: 20
device: h100
source_file_dependencies:
Expand All @@ -157,6 +162,7 @@ steps:

- label: ":nvidia: (B200) DeepSeek V4 Kernels"
key: deepseek-v4-kernel-test-b200
automatic_shard: true
timeout_in_minutes: 20
device: b200-k8s
source_file_dependencies:
Expand All @@ -173,13 +179,14 @@ steps:
- vllm/model_executor/layers/sparse_mqa_indexer.py
- tests/model_executor/layers/test_mla_short_prefill_indexer.py
commands:
- pytest -v -s kernels/test_fused_deepseek_v4_*.py
- pytest -v -s kernels/test_fused_deepseek_v4_qnorm_rope_kv_insert.py
- pytest -v -s kernels/moe/test_deep_gemm_mega_moe.py
- pytest -v -s models/test_deepseek_v4_mega_moe.py
# DeepSeek V4.1 sparse MQA logits (MXFP4 and FP8 indexer cache); SM100 only.
- pytest -v -s model_executor/layers/test_mla_short_prefill_indexer.py -k sparse_mqa_logits
mirror:
amd:
automatic_shard: true
label: ":amd: (MI355) DeepSeek V4 MoE Integration"
dind: false
device: mi355_1
Expand Down Expand Up @@ -212,6 +219,7 @@ steps:
# ptxas override.
- label: ":nvidia: (B200) mHC Kernels"
key: kernels-mhc-test-b200
automatic_shard: true
timeout_in_minutes: 25
device: b200-k8s
source_file_dependencies:
Expand All @@ -223,6 +231,7 @@ steps:
- pytest -v -s kernels/mhc
mirror:
amd:
automatic_shard: true
label: ":amd: (MI355) mHC Kernels"
dind: false
device: mi355_1
Expand All @@ -238,6 +247,7 @@ steps:

- label: ":nvidia: (B200) FLA Ops Kernels"
key: kernels-fla-ops-test-b200
automatic_shard: true
timeout_in_minutes: 25
device: b200-k8s
source_file_dependencies:
Expand All @@ -249,6 +259,7 @@ steps:
- pytest -v -s kernels/fla
mirror:
amd:
automatic_shard: true
label: ":amd: (MI355) FLA Ops Kernels"
dind: false
device: mi355_1
Expand All @@ -264,6 +275,7 @@ steps:

- label: ":nvidia: (B200) MiniMax M3 MegaMoE"
key: minimax-m3-mega-moe-test-b200
automatic_shard: true
timeout_in_minutes: 20
device: b200-k8s
source_file_dependencies:
Expand All @@ -276,6 +288,7 @@ steps:

- label: ":nvidia: (B200) MiniMax-M3 Model Tests"
key: minimax-m3-model-tests-b200
automatic_shard: true
timeout_in_minutes: 20
device: b200-k8s
source_file_dependencies:
Expand Down Expand Up @@ -410,6 +423,7 @@ steps:

- label: ":nvidia: (H100) Attention DiffKV Kernels"
key: kernels-attention-diffkv-test-h100
automatic_shard: true
timeout_in_minutes: 20
device: h100
num_devices: 1
Expand All @@ -422,6 +436,7 @@ steps:
- pytest -v -s kernels/attention/test_triton_unified_attention_diffkv.py
mirror:
amd:
automatic_shard: true
label: ":amd: (MI355 DPX) Attention DiffKV Kernels"
dind: false
device: mi355_dpx
Expand All @@ -441,6 +456,7 @@ steps:

- label: ":nvidia: (H100) MLA Kernel Test"
key: kernels-flashmla-test-h100
automatic_shard: true
timeout_in_minutes: 25
device: h100
num_devices: 1
Expand All @@ -462,6 +478,7 @@ steps:
- pytest -v -s kernels/attention/test_mla_cross_layer_kernel_equivalence.py
mirror:
amd:
automatic_shard: true
label: ":amd: (MI355 DPX) MLA Kernels"
dind: false
device: mi355_dpx
Expand Down Expand Up @@ -627,6 +644,7 @@ steps:

- label: ":nvidia: (B200) MiniMax-M3 Attention Kernels"
key: kernels-minimax-m3-attention-b200
automatic_shard: true
timeout_in_minutes: 30
working_dir: "/vllm-workspace/"
device: b200-k8s
Expand All @@ -641,6 +659,7 @@ steps:
- pytest -v -s tests/kernels/attention/test_minimax_m3.py
mirror:
amd:
automatic_shard: true
label: ":amd: (MI355 DPX) MiniMax-M3 Attention Kernels"
dind: false
device: mi355_dpx
Expand All @@ -654,6 +673,7 @@ steps:

- label: ":nvidia: (H100) MiniMax-M3 Sparse Prefill Kernels"
key: kernels-minimax-m3-sparse-prefill-h100
automatic_shard: true
timeout_in_minutes: 15
working_dir: "/vllm-workspace/"
device: h100
Expand Down Expand Up @@ -751,6 +771,7 @@ steps:
- pytest -v -s tests/kernels/mamba/test_gdn_prefill_cutedsl.py
mirror:
amd:
automatic_shard: true
label: ":amd: (MI355) NVFP4 Dense Emulation"
dind: false
device: mi355_1
Expand Down Expand Up @@ -778,6 +799,7 @@ steps:

- label: ":nvidia: (B200) Gate Linear CuTeDSL Kernels"
key: kernels-gate-linear-cutedsl-b200
automatic_shard: true
timeout_in_minutes: 15
working_dir: "/vllm-workspace/"
device: b200-k8s
Expand All @@ -799,6 +821,7 @@ steps:
- pytest -v -s tests/kernels/test_ll_fp32w_gemm.py
mirror:
amd:
automatic_shard: true
label: ":amd: (MI355) Gate Linear Router Kernels"
dind: false
device: mi355_1
Expand Down Expand Up @@ -889,6 +912,7 @@ steps:

- label: ":nvidia: (H100) [Nightly] FP8 MoE Kernels"
key: kernels-fp8-moe-test-1xh100
automatic_shard: true
timeout_in_minutes: 40
device: h100
num_devices: 1
Expand Down Expand Up @@ -937,6 +961,7 @@ steps:

- label: ":nvidia: (B200) [Nightly] FP4 MoE Kernels"
key: kernels-fp4-moe-test-b200
automatic_shard: true
timeout_in_minutes: 25
device: b200-k8s
num_devices: 1
Expand All @@ -948,6 +973,7 @@ steps:
- pytest -v -s kernels/moe/test_ocp_mx_moe.py
mirror:
amd:
automatic_shard: true
label: ":amd: (MI355 DPX) FP4 and INT4 MoE Kernels"
dind: false
device: mi355_dpx
Expand Down Expand Up @@ -1039,6 +1065,7 @@ steps:
- pytest -v -s kernels/moe/test_deepep_v2_moe.py
mirror:
amd:
automatic_shard: true
label: ":amd: (MI355) MoRI EP Numerics"
dind: false
device: mi355_2
Expand Down Expand Up @@ -1072,6 +1099,7 @@ steps:

- label: ":amd: (MI355 DPX) Native Quantization Kernels"
key: amd-native-quantization-kernels-mi355
automatic_shard: true
dind: false
device: mi355_dpx
num_devices: 1
Expand Down Expand Up @@ -1100,6 +1128,7 @@ steps:

- label: ":amd: (MI355 DPX) FP8 MoE Kernels"
key: amd-fp8-moe-kernels-mi355
automatic_shard: true
dind: false
device: mi355_dpx
num_devices: 1
Expand Down Expand Up @@ -1130,6 +1159,7 @@ steps:

- label: ":amd: (MI355 DPX) Kernels"
key: amd-kernels-mi355
automatic_shard: true
dind: false
device: mi355_dpx
num_devices: 1
Expand Down