[XPU][CI] Add more test cases in Intel GPU CI (#49422)
Signed-off-by: zengxian <xiangdong.zeng@intel.com> Co-authored-by: Kunshang Ji <kunshang.ji@intel.com>
xiangdong committed
Jul 27, 2026 at 12:40 UTC
29fdeab2548fdb5cd61ec3613e5e72200c695ef0
5 files changed
+219
-1
.buildkite/intel_jobs/benchmarks_intel.yaml
new
+26
@@ -0,0 +1,26 @@
1
+group: Benchmarks
2
+depends_on:
3
+ - image-build-xpu
4
+steps:
5
+- label: Benchmarks CLI Test
6
+ key: benchmarks-cli-test
7
+ timeout_in_minutes: 40
8
+ device: intel_gpu
9
+ agent_tags:
10
+ label: production
11
+ gpu: 1+
12
+ mem: 16+
13
+ no_plugin: true
14
+ working_dir: "."
15
+ env:
16
+ REGISTRY: "public.ecr.aws/q9t5s3a7"
17
+ REPO: "vllm-ci-test-repo"
18
+ VLLM_TEST_DEVICE: "xpu"
19
+ source_file_dependencies:
20
+ - vllm/
21
+ - tests/benchmarks/
22
+ commands:
23
+ - >-
24
+ bash .buildkite/scripts/hardware_ci/run-intel-test.sh
25
+ 'cd tests &&
26
+ pytest -v -s benchmarks/'
.buildkite/intel_jobs/engine_intel.yaml
+76
@@ -2,6 +2,44 @@ group: Engine Intel
2
depends_on:
3
- image-build-xpu
4
steps:
5
+- label: Engine
6
+ key: engine
7
+ timeout_in_minutes: 40
8
+ device: intel_gpu
9
+ agent_tags:
10
+ label: production
11
+ gpu: 1+
12
+ mem: 16+
13
+ no_plugin: true
14
+ working_dir: "."
15
+ env:
16
+ REGISTRY: "public.ecr.aws/q9t5s3a7"
17
+ REPO: "vllm-ci-test-repo"
18
+ VLLM_TEST_DEVICE: "xpu"
19
+ source_file_dependencies:
20
+ - vllm/compilation/
21
+ - vllm/config/
22
+ - vllm/engine/
23
+ - vllm/entrypoints/logger.py
24
+ - vllm/envs.py
25
+ - vllm/logger.py
26
+ - vllm/logging_utils/
27
+ - vllm/platforms/
28
+ - vllm/sequence.py
29
+ - vllm/triton_utils/
30
+ - vllm/utils/
31
+ - tests/engine
32
+ - tests/test_sequence
33
+ - tests/test_config
34
+ - tests/test_logger
35
+ - tests/test_vllm_port
36
+ - tests/test_jit_monitor.py
37
+ commands:
38
+ - >-
39
+ bash .buildkite/scripts/hardware_ci/run-intel-test.sh
40
+ 'cd tests &&
41
+ pytest -v -s engine/test_arg_utils.py test_sequence.py test_logger.py test_vllm_port.py test_jit_monitor.py'
42
+
43
- label: Engine (1 GPU)
44
timeout_in_minutes: 30
45
device: intel_gpu
@@ -23,3 +61,41 @@ steps:
61
bash .buildkite/scripts/hardware_ci/run-intel-test.sh
62
'cd tests &&
63
pytest -v -s v1/engine --ignore v1/engine/test_preprocess_error_handling.py'
64
+
65
+- label: V1 e2e (2 GPUs)
66
+ timeout_in_minutes: 30
67
+ device: intel_gpu
68
+ agent_tags:
69
+ label: production
70
+ gpu: 2+
71
+ mem: 16+
72
+ no_plugin: true
73
+ working_dir: "."
74
+ env:
75
+ REGISTRY: "public.ecr.aws/q9t5s3a7"
76
+ REPO: "vllm-ci-test-repo"
77
+ VLLM_TEST_DEVICE: "xpu"
78
+ source_file_dependencies:
79
+ - vllm/compilation/
80
+ - vllm/config/
81
+ - vllm/distributed/
82
+ - vllm/engine/
83
+ - vllm/envs.py
84
+ - vllm/forward_context.py
85
+ - vllm/inputs/
86
+ - vllm/logger.py
87
+ - vllm/logging_utils/
88
+ - vllm/model_executor/
89
+ - vllm/multimodal/
90
+ - vllm/platforms/
91
+ - vllm/sampling_params.py
92
+ - vllm/transformers_utils/
93
+ - vllm/triton_utils/
94
+ - vllm/utils/
95
+ - vllm/v1/
96
+ - tests/v1/e2e/spec_decode
97
+ commands:
98
+ - >-
99
+ bash .buildkite/scripts/hardware_ci/run-intel-test.sh
100
+ 'cd tests &&
101
+ pytest -v -s v1/e2e/spec_decode/test_spec_decode.py -k "tensor_parallelism"'
.buildkite/intel_jobs/model_executor_intel.yaml
new
+33
@@ -0,0 +1,33 @@
1
+group: Model Executor Intel
2
+depends_on:
3
+ - image-build-xpu
4
+steps:
5
+- label: Model Executor (Intel)
6
+ key: model-executor-intel
7
+ timeout_in_minutes: 45
8
+ device: intel_gpu
9
+ agent_tags:
10
+ label: production
11
+ gpu: 1+
12
+ mem: 24+
13
+ no_plugin: true
14
+ working_dir: "."
15
+ env:
16
+ REGISTRY: "public.ecr.aws/q9t5s3a7"
17
+ REPO: "vllm-ci-test-repo"
18
+ VLLM_TEST_DEVICE: "xpu"
19
+ source_file_dependencies:
20
+ - vllm/engine/arg_utils.py
21
+ - vllm/config/model.py
22
+ - vllm/model_executor
23
+ - tests/model_executor
24
+ commands:
25
+ - >-
26
+ bash .buildkite/scripts/hardware_ci/run-intel-test.sh
27
+ 'apt-get update && apt-get install -y curl libsodium23 &&
28
+ pip3 install tensorizer==2.10.1 &&
29
+ pip3 install runai-model-streamer[s3,gcs,azure]\>=0.15.7 &&
30
+ export VLLM_WORKER_MULTIPROC_METHOD=spawn &&
31
+ export PYTHONFAULTHANDLER=1 &&
32
+ cd tests &&
33
+ pytest -v -s model_executor -m "not slow_test" --ignore="model_executor/layers/test_rocm_unquantized_gemm.py" --deselect="tests/model_executor/model_loader/test_reload.py::test_kv_scale_reload"'
.buildkite/intel_jobs/model_runner_v2_intel.yaml
+55
-1
@@ -8,7 +8,7 @@ steps:
8
agent_tags:
9
label: production
10
gpu: 2+
11
- mem: 16+
11
+ mem: 24+
12
no_plugin: true
13
working_dir: "."
14
env:
@@ -28,7 +28,9 @@ steps:
28
'export VLLM_USE_V2_MODEL_RUNNER=1 &&
29
cd tests &&
30
pytest -v -s v1/engine/test_llm_engine.py -k "not test_engine_metrics" &&
31
+ pytest -v -s v1/e2e/general/test_context_length.py &&
32
ENFORCE_EAGER=1 pytest -v -s v1/e2e/general/test_async_scheduling.py -k "not ngram" &&
33
+ pytest -v -s entrypoints/llm/test_struct_output_generate.py -k "xgrammar and not speculative_config6 and not speculative_config7 and not speculative_config8 and not speculative_config0" &&
34
pytest -v -s v1/e2e/general/test_min_tokens.py'
35
36
- label: Model Runner V2 Examples (Intel)
@@ -60,3 +62,55 @@ steps:
62
python3 basic/offline_inference/generate.py --model facebook/opt-125m &&
63
python3 generate/multimodal/vision_language_offline.py --seed 0 &&
64
python3 features/automatic_prefix_caching/prefix_caching_offline.py'
65
+
66
+- label: Model Runner V2 Distributed (2 GPUs)
67
+ timeout_in_minutes: 50
68
+ device: intel_gpu
69
+ agent_tags:
70
+ label: production
71
+ gpu: 2+
72
+ mem: 16+
73
+ no_plugin: true
74
+ working_dir: "."
75
+ env:
76
+ REGISTRY: "public.ecr.aws/q9t5s3a7"
77
+ REPO: "vllm-ci-test-repo"
78
+ VLLM_TEST_DEVICE: "xpu"
79
+ source_file_dependencies:
80
+ - vllm/v1/worker/gpu/
81
+ - vllm/v1/worker/gpu_worker.py
82
+ - tests/basic_correctness/test_basic_correctness.py
83
+ - tests/v1/distributed/test_async_llm_dp.py
84
+ - tests/v1/distributed/test_eagle_dp.py
85
+ commands:
86
+ - >-
87
+ bash .buildkite/scripts/hardware_ci/run-intel-test.sh
88
+ 'export VLLM_USE_V2_MODEL_RUNNER=1 &&
89
+ cd tests &&
90
+ TARGET_TEST_SUITE=L4 pytest -v -s basic_correctness/test_basic_correctness.py -m "distributed\(num_gpus=2\)" -k "not ray and not True"'
91
+
92
+- label: Model Runner V2 Spec Decode
93
+ timeout_in_minutes: 50
94
+ device: intel_gpu
95
+ agent_tags:
96
+ label: production
97
+ gpu: 1+
98
+ mem: 24+
99
+ no_plugin: true
100
+ working_dir: "."
101
+ env:
102
+ REGISTRY: "public.ecr.aws/q9t5s3a7"
103
+ REPO: "vllm-ci-test-repo"
104
+ VLLM_TEST_DEVICE: "xpu"
105
+ source_file_dependencies:
106
+ - vllm/v1/worker/gpu/
107
+ - vllm/v1/worker/gpu_worker.py
108
+ - tests/v1/spec_decode/test_max_len.py
109
+ - tests/v1/spec_decode/test_rejection_sampler_utils.py
110
+ - tests/v1/e2e/spec_decode/test_spec_decode.py
111
+ commands:
112
+ - >-
113
+ bash .buildkite/scripts/hardware_ci/run-intel-test.sh
114
+ 'export VLLM_USE_V2_MODEL_RUNNER=1 &&
115
+ cd tests &&
116
+ pytest -v -s v1/spec_decode/test_synthetic_rejection_sampler_utils.py'
.buildkite/intel_jobs/samplers_intel.yaml
new
+29
@@ -0,0 +1,29 @@
1
+group: Samplers Intel
2
+depends_on:
3
+ - image-build-xpu
4
+steps:
5
+- label: Samplers Test (FlashInfer)
6
+ key: samplers-test-flashinfer-intel
7
+ timeout_in_minutes: 40
8
+ device: intel_gpu
9
+ agent_tags:
10
+ label: production
11
+ gpu: 1+
12
+ mem: 24+
13
+ no_plugin: true
14
+ working_dir: "."
15
+ env:
16
+ REGISTRY: "public.ecr.aws/q9t5s3a7"
17
+ REPO: "vllm-ci-test-repo"
18
+ VLLM_TEST_DEVICE: "xpu"
19
+ source_file_dependencies:
20
+ - vllm/model_executor/layers
21
+ - vllm/sampling_metadata.py
22
+ - tests/samplers
23
+ - tests/conftest.py
24
+ - vllm/entrypoints/generate/beam_search
25
+ commands:
26
+ - >-
27
+ bash .buildkite/scripts/hardware_ci/run-intel-test.sh
28
+ 'cd tests &&
29
+ VLLM_USE_FLASHINFER_SAMPLER=1 pytest -v -s samplers'