[CI] Shard H100 MoE refactor integration tests (#52352)
Signed-off-by: Kevin Luu <51931015+khluu@users.noreply.github.com>
Kevin H. Luu committed
Sep 2, 2026 at 01:10 UTC
1b4b2a18b80c5e265b93ab6f83dde494eadfaa4a
6 files changed
+17
-13
.buildkite/test_areas/lm_eval.yaml
+5
-1
@@ -177,8 +177,12 @@ steps:
177
device: h100
178
optional: true
179
num_devices: 2
180
+ parallelism: 4
181
commands:
181
- - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/moe-refactor/config-h100.txt
182
+ # Timing-balanced static split of the former config-h100.txt (measured on
183
+ # build 83851): shards ~12.4/13.6/12.4/14.5m of eval time. New configs must
184
+ # be added to exactly one config-h100-shard-*.txt.
185
+ - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/moe-refactor/config-h100-shard-$$BUILDKITE_PARALLEL_JOB.txt
186
187
- label: ":nvidia: (B200) MoE Refactor Integration TEMPORARY Shard %N"
188
key: moe-refactor-integration-test-b200-temporary
tests/evals/gsm8k/configs/moe-refactor/config-h100-shard-0.txt
new
+2
@@ -0,0 +1,2 @@
1
+Qwen3-30B-A3B-Fp8-AutoFp8-deepgemm.yaml
2
+Qwen3-30B-A3B-BF16-fi-cutlass.yaml
tests/evals/gsm8k/configs/moe-refactor/config-h100-shard-1.txt
new
+3
@@ -0,0 +1,3 @@
1
+Qwen3-30B-A3B-Fp8-AutoFp8-marlin.yaml
2
+Qwen3-30B-A3B-Fp8-AutoFp8-triton.yaml
3
+Qwen3-30B-A3B-BF16-triton.yaml
tests/evals/gsm8k/configs/moe-refactor/config-h100-shard-2.txt
new
+3
@@ -0,0 +1,3 @@
1
+Qwen3-30B-A3B-Fp8-AutoFp8-fi-cutlass.yaml
2
+Qwen3-30B-A3B-Fp8-CT-Block-deepgemm.yaml
3
+Qwen3-30B-A3B-Fp8-CT-Block-triton.yaml
tests/evals/gsm8k/configs/moe-refactor/config-h100-shard-3.txt
new
+4
@@ -0,0 +1,4 @@
1
+Mixtral-8x7B-Fp8-AutoFp8-triton.yaml
2
+Qwen3-30B-A3B-Fp8-CT-Block-marlin.yaml
3
+Qwen3-30B-A3B-Fp8-CT-Channel-marlin.yaml
4
+Qwen3-30B-A3B-Fp8-CT-Channel-vllm-cutlass.yaml
tests/evals/gsm8k/configs/moe-refactor/config-h100.txt
deleted
-12
@@ -1,12 +0,0 @@
1
-Mixtral-8x7B-Fp8-AutoFp8-triton.yaml
2
-Qwen3-30B-A3B-Fp8-AutoFp8-deepgemm.yaml
3
-Qwen3-30B-A3B-Fp8-AutoFp8-fi-cutlass.yaml
4
-Qwen3-30B-A3B-Fp8-AutoFp8-marlin.yaml
5
-Qwen3-30B-A3B-Fp8-AutoFp8-triton.yaml
6
-Qwen3-30B-A3B-Fp8-CT-Block-deepgemm.yaml
7
-Qwen3-30B-A3B-Fp8-CT-Block-marlin.yaml
8
-Qwen3-30B-A3B-Fp8-CT-Block-triton.yaml
9
-Qwen3-30B-A3B-Fp8-CT-Channel-marlin.yaml
10
-Qwen3-30B-A3B-Fp8-CT-Channel-vllm-cutlass.yaml
11
-Qwen3-30B-A3B-BF16-fi-cutlass.yaml
12
-Qwen3-30B-A3B-BF16-triton.yaml