[ROCm][CI] Stage C-II of gating additional test groups (#46537)
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Andreas Karatzas committed
Jun 23, 2026 at 19:36 UTC
bcbeaac786c1b86076ad2c161eca4c34693e4dcc
10 files changed
+96
-15
.buildkite/test-amd.yaml
+5
-15
@@ -682,6 +682,7 @@ steps:
682
timeout_in_minutes: 180
683
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
684
agent_pool: mi300_1
685
+ optional: true
686
working_dir: "/vllm-workspace/tests"
687
source_file_dependencies:
688
- vllm/
@@ -829,6 +830,7 @@ steps:
830
timeout_in_minutes: 180
831
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
832
agent_pool: mi300_1
833
+ optional: true
834
working_dir: "/vllm-workspace/tests"
835
source_file_dependencies:
836
- vllm/distributed/eplb
@@ -1631,6 +1633,7 @@ steps:
1633
timeout_in_minutes: 180
1634
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1635
agent_pool: mi300_1
1636
+ optional: true
1637
torch_nightly: true
1638
parallelism: 2
1639
working_dir: "/vllm-workspace/tests"
@@ -2197,6 +2200,7 @@ steps:
2200
timeout_in_minutes: 180
2201
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
2202
agent_pool: mi300_2
2203
+ optional: true
2204
num_gpus: 2
2205
working_dir: "/vllm-workspace/tests"
2206
source_file_dependencies:
@@ -2215,21 +2219,6 @@ steps:
2219
- TP_SIZE=1 DP_SIZE=2 pytest -v -s v1/distributed/test_external_lb_dp.py
2220
- DP_SIZE=2 pytest -v -s entrypoints/openai/test_multi_api_servers.py
2221
2218
-- label: NixlConnector PD + Spec Decode acceptance (2 GPUs) # TBD
2219
- timeout_in_minutes: 180
2220
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
2221
- agent_pool: mi300_2
2222
- num_gpus: 2
2223
- working_dir: "/vllm-workspace/tests"
2224
- source_file_dependencies:
2225
- - vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py
2226
- - vllm/v1/worker/kv_connector_model_runner_mixin.py
2227
- - tests/v1/kv_connector/nixl_integration/
2228
- - vllm/platforms/rocm.py
2229
- commands:
2230
- - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt
2231
- - ATTENTION_BACKEND=TRITON_ATTN bash v1/kv_connector/nixl_integration/spec_decode_acceptance_test.sh
2232
-
2222
- label: Distributed Tests (2xH100-2xMI300) # TBD
2223
timeout_in_minutes: 180
2224
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
@@ -2254,6 +2243,7 @@ steps:
2243
timeout_in_minutes: 180
2244
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
2245
agent_pool: mi300_2
2246
+ optional: true
2247
num_gpus: 2
2248
working_dir: "/vllm-workspace/tests"
2249
source_file_dependencies:
.buildkite/test_areas/benchmarks.yaml
+5
@@ -11,6 +11,11 @@ steps:
11
- tests/benchmarks/
12
commands:
13
- pytest -v -s benchmarks/
14
+ mirror:
15
+ amd:
16
+ device: mi300_1
17
+ depends_on:
18
+ - image-build-amd
19
20
- label: Attention Benchmarks Smoke Test (B200)
21
key: attention-benchmarks-smoke-test-b200
.buildkite/test_areas/distributed.yaml
+15
@@ -37,6 +37,21 @@ steps:
37
- TP_SIZE=1 DP_SIZE=2 pytest -v -s v1/distributed/test_eagle_dp.py
38
- TP_SIZE=1 DP_SIZE=2 pytest -v -s v1/distributed/test_external_lb_dp.py
39
- DP_SIZE=2 pytest -v -s entrypoints/openai/test_multi_api_servers.py
40
+ mirror:
41
+ amd:
42
+ device: mi300_2
43
+ depends_on:
44
+ - image-build-amd
45
+ source_file_dependencies:
46
+ - vllm/distributed/
47
+ - vllm/engine/
48
+ - vllm/executor/
49
+ - vllm/worker/worker_base.py
50
+ - vllm/v1/engine/
51
+ - vllm/v1/worker/
52
+ - tests/v1/distributed
53
+ - tests/entrypoints/openai/test_multi_api_servers.py
54
+ - vllm/platforms/rocm.py
55
56
- label: Distributed Compile + RPC Tests (2 GPUs)
57
key: distributed-compile-rpc-tests-2-gpus
.buildkite/test_areas/engine.yaml
+5
@@ -112,6 +112,11 @@ steps:
112
commands:
113
# Only run tests that need exactly 2 GPUs
114
- pytest -v -s v1/e2e/spec_decode/test_spec_decode.py -k "tensor_parallelism"
115
+ mirror:
116
+ amd:
117
+ device: mi300_2
118
+ depends_on:
119
+ - image-build-amd
120
121
- label: V1 e2e (4 GPUs)
122
key: v1-e2e-4-gpus
.buildkite/test_areas/expert_parallelism.yaml
+10
@@ -14,6 +14,16 @@ steps:
14
commands:
15
- pytest -v -s distributed/test_eplb_algo.py
16
- pytest -v -s distributed/test_eplb_utils.py
17
+ mirror:
18
+ amd:
19
+ device: mi300_1
20
+ depends_on:
21
+ - image-build-amd
22
+ source_file_dependencies:
23
+ - vllm/distributed/eplb
24
+ - tests/distributed/test_eplb_algo.py
25
+ - tests/distributed/test_eplb_utils.py
26
+ - vllm/platforms/rocm.py
27
28
- label: EPLB Execution # 17min
29
key: eplb-execution
.buildkite/test_areas/misc.yaml
+5
@@ -250,6 +250,11 @@ steps:
250
'opentelemetry-exporter-otlp>=1.26.0' \
251
'opentelemetry-semantic-conventions-ai>=0.4.1'"
252
- pytest -v -s v1/tracing
253
+ mirror:
254
+ amd:
255
+ device: mi300_2
256
+ depends_on:
257
+ - image-build-amd
258
259
- label: Python-only Installation
260
key: python-only-installation
.buildkite/test_areas/model_executor.yaml
+13
@@ -23,3 +23,16 @@ steps:
23
# calls that the signal method cannot interrupt.
24
- pytest -v -s model_executor -m '(not slow_test)' --timeout=900 --timeout-method=thread
25
- pytest -v -s entrypoints/openai/completion/test_tensorizer_entrypoint.py --timeout=900 --timeout-method=thread
26
+ mirror:
27
+ amd:
28
+ device: mi300_1
29
+ depends_on:
30
+ - image-build-amd
31
+ source_file_dependencies:
32
+ - vllm/engine/arg_utils.py
33
+ - vllm/config/model.py
34
+ - vllm/model_executor
35
+ - tests/model_executor
36
+ - tests/entrypoints/openai/completion/test_tensorizer_entrypoint.py
37
+ - vllm/_aiter_ops.py
38
+ - vllm/platforms/rocm.py
.buildkite/test_areas/models_language.yaml
+19
@@ -15,6 +15,10 @@ steps:
15
- pytest -v -s models/language -m 'core_model and (not slow_test)'
16
mirror:
17
torch_nightly: {}
18
+ amd:
19
+ device: mi300_1
20
+ depends_on:
21
+ - image-build-amd
22
23
- label: Language Models Tests (Extra Standard) %N
24
key: language-models-tests-extra-standard
@@ -32,6 +36,21 @@ steps:
36
parallelism: 2
37
mirror:
38
torch_nightly: {}
39
+ amd:
40
+ device: mi300_1
41
+ depends_on:
42
+ - image-build-amd
43
+ source_file_dependencies:
44
+ - vllm/model_executor/models/
45
+ - vllm/model_executor/model_loader/
46
+ - vllm/model_executor/layers/
47
+ - vllm/v1/attention/backends/
48
+ - vllm/v1/attention/selector.py
49
+ - tests/models/language/pooling/test_embedding.py
50
+ - tests/models/language/generation/test_common.py
51
+ - tests/models/language/pooling/test_classification.py
52
+ - vllm/_aiter_ops.py
53
+ - vllm/platforms/rocm.py
54
55
- label: Language Models Tests (Hybrid) %N
56
key: language-models-tests-hybrid
.buildkite/test_areas/models_multimodal.yaml
+11
@@ -109,6 +109,17 @@ steps:
109
- vllm/v1/core/
110
commands:
111
- pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-mm-small.txt --tp-size=1
112
+ mirror:
113
+ amd:
114
+ device: mi300_1
115
+ depends_on:
116
+ - image-build-amd
117
+ source_file_dependencies:
118
+ - vllm/multimodal/
119
+ - vllm/inputs/
120
+ - vllm/v1/core/
121
+ - vllm/platforms/rocm.py
122
+ - vllm/model_executor/model_loader/
123
124
- label: Multi-Modal Models (Extended Generation 1)
125
key: multi-modal-models-extended-generation-1
.buildkite/test_areas/pytorch.yaml
+8
@@ -195,3 +195,11 @@ steps:
195
- requirements/test/nightly-torch.txt
196
commands:
197
- bash standalone_tests/pytorch_nightly_dependency.sh
198
+ mirror:
199
+ amd:
200
+ device: mi300_1
201
+ depends_on:
202
+ - image-build-amd
203
+ source_file_dependencies:
204
+ - requirements/test/nightly-torch.txt
205
+ - vllm/platforms/rocm.py