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 3, 14:04 UTC
[ROCm][CI] Stage B gating (#42025)
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Andreas Karatzas committed
May 15, 2026 at 03:49 UTC
d735968f6d634ec849268f18e3b84ceb494fee79
5 files changed
+61
-27
.buildkite/test-amd.yaml
+6
-26
@@ -798,32 +798,6 @@ steps:
798
commands:
799
- pytest -v -s v1/attention
800
801
-- label: V1 Core + KV + Metrics # TBD
802
- timeout_in_minutes: 60
803
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250]
804
- agent_pool: mi250_1
805
- optional: true
806
- working_dir: "/vllm-workspace/tests"
807
- source_file_dependencies:
808
- - vllm/
809
- - tests/v1/core
810
- - tests/v1/executor
811
- - tests/v1/kv_offload
812
- - tests/v1/worker
813
- - tests/v1/kv_connector/unit
814
- - tests/v1/metrics
815
- - tests/entrypoints/openai/correctness/test_lmeval.py
816
- commands:
817
- - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt
818
- - pytest -v -s -m 'not cpu_test' v1/core
819
- - pytest -v -s v1/executor
820
- - pytest -v -s v1/kv_offload
821
- - pytest -v -s v1/worker
822
- - pytest -v -s -m 'not cpu_test' v1/kv_connector/unit
823
- - pytest -v -s -m 'not cpu_test' v1/metrics
824
- - pip install -U git+https://github.com/robertgshaw2-redhat/lm-evaluation-harness.git@streaming-api
825
- - pytest -v -s entrypoints/openai/correctness/test_lmeval.py::test_lm_eval_accuracy_v1_engine
826
-
801
- label: V1 Sample + Logits # TBD
802
timeout_in_minutes: 60
803
mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250]
@@ -1279,6 +1253,7 @@ steps:
1253
timeout_in_minutes: 180
1254
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1255
agent_pool: mi300_1
1256
+ optional: true
1257
fast_check: true
1258
torch_nightly: true
1259
working_dir: "/vllm-workspace/tests"
@@ -1299,6 +1274,7 @@ steps:
1274
agent_pool: mi300_1
1275
fast_check: true
1276
torch_nightly: true
1277
+ optional: true
1278
working_dir: "/vllm-workspace/tests"
1279
source_file_dependencies:
1280
- vllm/
@@ -1312,6 +1288,7 @@ steps:
1288
timeout_in_minutes: 180
1289
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1290
agent_pool: mi300_1
1291
+ optional: true
1292
fast_check: true
1293
torch_nightly: true
1294
working_dir: "/vllm-workspace/tests"
@@ -1328,6 +1305,7 @@ steps:
1305
timeout_in_minutes: 180
1306
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1307
agent_pool: mi300_1
1308
+ optional: true
1309
fast_check: true
1310
torch_nightly: true
1311
working_dir: "/vllm-workspace/tests"
@@ -1728,6 +1706,7 @@ steps:
1706
timeout_in_minutes: 180
1707
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1708
agent_pool: mi300_1
1709
+ optional: true
1710
parallelism: 2
1711
working_dir: "/vllm-workspace/tests"
1712
source_file_dependencies:
@@ -1786,6 +1765,7 @@ steps:
1765
timeout_in_minutes: 180
1766
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1767
agent_pool: mi300_1
1768
+ optional: true
1769
working_dir: "/vllm-workspace/tests"
1770
source_file_dependencies:
1771
- vllm/
.buildkite/test_areas/engine.yaml
+12
@@ -36,6 +36,12 @@ steps:
36
commands:
37
- pytest -v -s v1/engine/test_preprocess_error_handling.py
38
- pytest -v -s v1/engine --ignore v1/engine/test_preprocess_error_handling.py
39
+ mirror:
40
+ amd:
41
+ device: mi300_1
42
+ timeout_in_minutes: 40
43
+ depends_on:
44
+ - image-build-amd
45
46
- label: e2e Scheduling (1 GPU)
47
key: e2e-scheduling-1-gpu
@@ -46,6 +52,12 @@ steps:
52
- tests/v1/e2e/general/
53
commands:
54
- pytest -v -s v1/e2e/general/test_async_scheduling.py
55
+ mirror:
56
+ amd:
57
+ device: mi250_1
58
+ timeout_in_minutes: 60
59
+ depends_on:
60
+ - image-build-amd
61
62
- label: e2e Core (1 GPU)
63
key: e2e-core-1-gpu
.buildkite/test_areas/entrypoints.yaml
+24
-1
@@ -43,6 +43,12 @@ steps:
43
commands:
44
- export VLLM_WORKER_MULTIPROC_METHOD=spawn
45
- pytest -v -s entrypoints/openai/chat_completion --ignore=entrypoints/openai/chat_completion/test_chat_with_tool_reasoning.py --ignore=entrypoints/openai/chat_completion/test_oot_registration.py
46
+ mirror:
47
+ amd:
48
+ device: mi300_1
49
+ timeout_in_minutes: 80
50
+ depends_on:
51
+ - image-build-amd
52
53
- label: Entrypoints Integration (API Server openai - Part 2)
54
key: entrypoints-integration-api-server-openai-part-2
@@ -55,6 +61,12 @@ steps:
61
commands:
62
- pytest -v -s entrypoints/openai/completion --ignore=entrypoints/openai/completion/test_tensorizer_entrypoint.py
63
- pytest -v -s entrypoints/test_chat_utils.py
64
+ mirror:
65
+ amd:
66
+ device: mi300_1
67
+ timeout_in_minutes: 60
68
+ depends_on:
69
+ - image-build-amd
70
71
- label: Entrypoints Integration (API Server openai - Part 3)
72
key: entrypoints-integration-api-server-openai-part-3
@@ -67,7 +79,13 @@ steps:
79
- tests/entrypoints/test_chat_utils
80
commands:
81
- export VLLM_WORKER_MULTIPROC_METHOD=spawn
70
- - pytest -v -s entrypoints/openai --ignore=entrypoints/openai/chat_completion --ignore=entrypoints/openai/completion --ignore=entrypoints/openai/correctness/ --ignore=entrypoints/openai/tool_parsers/ --ignore=entrypoints/openai/responses --ignore=entrypoints/openai/test_multi_api_servers.py
82
+ - pytest -v -s entrypoints/openai --ignore=entrypoints/openai/chat_completion --ignore=entrypoints/openai/completion --ignore=entrypoints/openai/correctness/ --ignore=entrypoints/openai/tool_parsers/ --ignore=entrypoints/openai/responses --ignore=entrypoints/openai/test_multi_api_servers.py
83
+ mirror:
84
+ amd:
85
+ device: mi300_1
86
+ timeout_in_minutes: 60
87
+ depends_on:
88
+ - image-build-amd
89
90
- label: Entrypoints Integration (API Server 2)
91
key: entrypoints-integration-api-server-2
@@ -83,6 +101,11 @@ steps:
101
- pytest -v -s entrypoints/serve/instrumentator
102
- PYTHONPATH=/vllm-workspace pytest -v -s entrypoints/rpc
103
- pytest -v -s tool_use
104
+ mirror:
105
+ amd:
106
+ device: mi300_1
107
+ depends_on:
108
+ - image-build-amd
109
110
- label: Entrypoints Integration (Speech to Text)
111
key: entrypoints-integration-speech_to_text
.buildkite/test_areas/kernels.yaml
+13
@@ -62,6 +62,19 @@ steps:
62
commands:
63
- pytest -v -s kernels/quantization --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
64
parallelism: 2
65
+ mirror:
66
+ amd:
67
+ device: mi300_1
68
+ source_file_dependencies:
69
+ - csrc/quantization/
70
+ - vllm/model_executor/layers/quantization
71
+ - tests/kernels/quantization
72
+ - tests/kernels/quantization/test_rocm_skinny_gemms.py
73
+ - vllm/_aiter_ops.py
74
+ - vllm/platforms/rocm.py
75
+ - vllm/model_executor/kernels/
76
+ depends_on:
77
+ - image-build-amd
78
79
- label: Kernels MoE Test %N
80
key: kernels-moe-test
.buildkite/test_areas/models_language.yaml
+6
@@ -92,6 +92,12 @@ steps:
92
- tests/models/language/pooling
93
commands:
94
- pytest -v -s models/language/pooling -m 'not core_model'
95
+ mirror:
96
+ amd:
97
+ device: mi300_1
98
+ timeout_in_minutes: 100
99
+ depends_on:
100
+ - image-build-amd
101
102
- label: Language Models Test (MTEB)
103
key: language-models-test-mteb