[CI][XPU] Disable model runner V2 for XPU quantization test for some partially pre-quantized models (#56179)
Signed-off-by: Chaojun Zhang <chaojun.zhang@intel.com>
Chaojun Zhang committed
Sep 10, 2026 at 11:02 UTC
5c0aeb5954be3384ea17ebdb8bb21719f538492a
1 file changed
+6
.buildkite/intel_jobs/test-intel.yaml
+6
@@ -145,10 +145,16 @@ steps:
145
source_file_dependencies:
146
- vllm/
147
- .buildkite/intel_jobs/test-intel.yaml
148
+ # TODO: XPU fused MoE + model runner V2 currently hangs on some
149
+ # pre-quantized models, e.g.
150
+ # quantization/test_online.py::test_online_quantization_loads_real_weights[nm-testing/tinysmokeqwen3moe-W4A16-first-only-CTstable]
151
+ # Remove the VLLM_USE_V2_MODEL_RUNNER=0 override below once the hang is
152
+ # root-caused and fixed.
153
commands:
154
- >-
155
bash .buildkite/scripts/hardware_ci/run-intel-test.sh
156
'cd tests &&
157
+ export VLLM_USE_V2_MODEL_RUNNER=0 &&
158
pytest -v -s quantization/test_auto_round.py &&
159
pytest -v -s quantization/test_online.py'
160
- label: "XPU compressed tensors FP8 test"