[CI] Run DFlash2 NVFP4 acceptance test on B200; skip it on H200 35GB MIG (#58496)
Signed-off-by: khluu <khluu000@gmail.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: Misha Goin <mgoin64@gmail.com>
Kevin H. Luu committed
Sep 24, 2026 at 18:29 UTC
155b988373cbe196e0b018ad052c1ee50d83311b
1 file changed
+17
-1
.buildkite/test_areas/spec_decode.yaml
+17
-1
@@ -290,7 +290,8 @@ steps:
290
- vllm/model_executor/models/laguna_dflash.py
291
- tests/v1/e2e/spec_decode/
292
commands:
293
- - pytest -v -s v1/e2e/spec_decode/acceptance_rates/dflash/
293
+ # DFlash2 NVFP4 (27B) does not fit a 35GB MIG slice; runs on B200 below.
294
+ - pytest -v -s v1/e2e/spec_decode/acceptance_rates/dflash/ -k "not dflash2"
295
mirror:
296
amd:
297
label: ":amd: (MI300) Spec Decode AL DFlash Nightly"
@@ -316,6 +317,21 @@ steps:
317
- tests/v1/e2e/spec_decode/
318
- vllm/platforms/rocm.py
319
320
+- label: ":nvidia: (B200) Spec Decode AL DFlash2 Nightly"
321
+ key: spec-decode-dflash2-nightly-b200
322
+ timeout_in_minutes: 60
323
+ device: b200-k8s
324
+ optional: true
325
+ source_file_dependencies:
326
+ - vllm/v1/spec_decode/
327
+ - vllm/v1/worker/gpu/spec_decode/
328
+ - vllm/model_executor/models/qwen3_dflash.py
329
+ - vllm/model_executor/models/qwen3_dflash2.py
330
+ - vllm/model_executor/models/qwen3_5.py
331
+ - tests/v1/e2e/spec_decode/
332
+ commands:
333
+ - pytest -v -s v1/e2e/spec_decode/acceptance_rates/dflash/ -k "dflash2"
334
+
335
- label: ":nvidia: (H200 MIG 35GB) Spec Decode AL DSpark Nightly"
336
key: spec-decode-dspark-nightly
337
timeout_in_minutes: 60