[CI] Add DGX Spark GPQA smoke test (#39541)
Signed-off-by: mgoin <mgoin64@gmail.com> Signed-off-by: Michael Goin <mgoin64@gmail.com> Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Michael Goin committed
Jun 22, 2026 at 22:52 UTC
6cc2c9ba3a2c78714b897474e087745058570902
4 files changed
+24
-2
.buildkite/image_build/image_build_arm64.sh
+3
-2
@@ -21,12 +21,13 @@ else
21
exit 0
22
fi
23
24
-# build (Grace/GH200 is the arm64 GPU target; sm_90)
24
+# build for arm64 GPU targets: Grace/GH200 (sm_90) and DGX Spark/GB10
25
+# (sm_121, family-covered by 12.0 under CUDA 13)
26
docker build --file docker/Dockerfile \
27
--platform linux/arm64 \
28
--build-arg max_jobs=16 \
29
--build-arg nvcc_threads=4 \
29
- --build-arg torch_cuda_arch_list="9.0" \
30
+ --build-arg torch_cuda_arch_list="9.0 12.0" \
31
--build-arg USE_SCCACHE=1 \
32
--build-arg buildkite_commit="$BUILDKITE_COMMIT" \
33
--tag "$REGISTRY"/"$REPO":"$BUILDKITE_COMMIT"-arm64 \
.buildkite/test_areas/lm_eval.yaml
+14
@@ -220,6 +220,20 @@ steps:
220
- uv pip install --system 'gpt-oss[eval]==0.0.5'
221
- pytest -s -v evals/gpt_oss/test_gpqa_correctness.py --config-list-file=configs/models-b200.txt
222
223
+- label: GPQA Eval (GPT-OSS) (DGX Spark)
224
+ key: gpqa-eval-gpt-oss-spark
225
+ timeout_in_minutes: 120
226
+ device: dgx-spark
227
+ optional: true
228
+ num_devices: 1
229
+ source_file_dependencies:
230
+ - csrc/
231
+ - vllm/model_executor/layers/quantization
232
+ - tests/evals/gpt_oss/
233
+ commands:
234
+ - uv pip install --system 'gpt-oss[eval]==0.0.5'
235
+ - pytest -s -v evals/gpt_oss/test_gpqa_correctness.py --config-list-file=configs/models-spark.txt
236
+
237
- label: MRCR Eval Small Models
238
device: h200_35gb
239
timeout_in_minutes: 30
tests/evals/gpt_oss/configs/gpt-oss-20b-sm120.yaml
new
+5
@@ -0,0 +1,5 @@
1
+# SPDX-License-Identifier: Apache-2.0
2
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
3
+model_name: "openai/gpt-oss-20b"
4
+metric_threshold: 0.568
5
+reasoning_effort: "low"
tests/evals/gpt_oss/configs/models-spark.txt
new
+2
@@ -0,0 +1,2 @@
1
+# DGX Spark model configurations for GPQA evaluation
2
+gpt-oss-20b-sm120.yaml