[CI] Split nightly MTP acceptance tests (#52353)
Signed-off-by: Kevin Luu <51931015+khluu@users.noreply.github.com>
Kevin H. Luu committed
Sep 1, 2026 at 15:29 UTC
e90b608c42f60b03c20f57fabe22ec0d3cacd76c
1 file changed
+8
-2
.buildkite/test_areas/spec_decode.yaml
+8
-2
@@ -349,13 +349,19 @@ steps:
349
350
- label: ":nvidia: (H200 MIG 35GB) Spec Decode AL MTP + Other Acceptance Nightly"
351
key: spec-decode-mtp-other-acceptance-nightly
352
- timeout_in_minutes: 60
352
+ timeout_in_minutes: 35
353
device: h200_35gb
354
optional: true
355
+ parallelism: 3
356
source_file_dependencies:
357
- vllm/v1/spec_decode/
358
- vllm/v1/worker/gpu/spec_decode/
359
- tests/v1/e2e/spec_decode/conftest.py
360
- tests/v1/e2e/spec_decode/
361
commands:
361
- - pytest -v -s v1/e2e/spec_decode/acceptance_rates/mtp_other/
362
+ # Each Gemma4 parametrization is an indivisible ~23m floor. Keep them in
363
+ # separate shards and put the remaining tests in shard 0, where new tests
364
+ # land by default.
365
+ - test "$$BUILDKITE_PARALLEL_JOB" != "0" || pytest -v -s v1/e2e/spec_decode/acceptance_rates/mtp_other/ --deselect "tests/v1/e2e/spec_decode/acceptance_rates/mtp_other/test_mtp.py::test_gemma4_mtp_acceptance_lengths[False]" --deselect "tests/v1/e2e/spec_decode/acceptance_rates/mtp_other/test_mtp.py::test_gemma4_mtp_acceptance_lengths[True]"
366
+ - test "$$BUILDKITE_PARALLEL_JOB" != "1" || pytest -v -s "v1/e2e/spec_decode/acceptance_rates/mtp_other/test_mtp.py::test_gemma4_mtp_acceptance_lengths[False]"
367
+ - test "$$BUILDKITE_PARALLEL_JOB" != "2" || pytest -v -s "v1/e2e/spec_decode/acceptance_rates/mtp_other/test_mtp.py::test_gemma4_mtp_acceptance_lengths[True]"