A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 5, 23:30 UTC
[CI] Add batch invariant test for b200 (#38014)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Wentao Ye committed
Mar 26, 2026 at 11:54 UTC
e054f152faa48ab27389f490d6e86c959d86d122
1 file changed
+15
.buildkite/test_areas/misc.yaml
+15
@@ -205,6 +205,21 @@ steps:
205
- pytest -v -s v1/determinism/test_rms_norm_batch_invariant.py
206
- VLLM_TEST_MODEL=deepseek-ai/DeepSeek-V2-Lite-Chat pytest -v -s v1/determinism/test_batch_invariance.py::test_v1_generation_is_deterministic_across_batch_sizes_with_needle[TRITON_MLA]
207
- VLLM_TEST_MODEL=Qwen/Qwen3-30B-A3B-Thinking-2507-FP8 pytest -v -s v1/determinism/test_batch_invariance.py::test_v1_generation_is_deterministic_across_batch_sizes_with_needle[FLASH_ATTN]
208
+
209
+- label: Batch Invariance (B200)
210
+ timeout_in_minutes: 30
211
+ device: b200
212
+ source_file_dependencies:
213
+ - vllm/v1/attention
214
+ - vllm/model_executor/layers
215
+ - tests/v1/determinism/
216
+ commands:
217
+ - export VLLM_WORKER_MULTIPROC_METHOD=spawn
218
+ - pip install pytest-timeout pytest-forked
219
+ - pytest -v -s v1/determinism/test_batch_invariance.py
220
+ - pytest -v -s v1/determinism/test_rms_norm_batch_invariant.py
221
+ - VLLM_TEST_MODEL=deepseek-ai/DeepSeek-V2-Lite-Chat pytest -v -s v1/determinism/test_batch_invariance.py::test_v1_generation_is_deterministic_across_batch_sizes_with_needle[TRITON_MLA]
222
+ - VLLM_TEST_MODEL=Qwen/Qwen3-30B-A3B-Thinking-2507-FP8 pytest -v -s v1/determinism/test_batch_invariance.py::test_v1_generation_is_deterministic_across_batch_sizes_with_needle[FLASH_ATTN]
223
224
- label: Acceptance Length Test (Large Models) # optional
225
timeout_in_minutes: 25