[ROCm][CI] Gating more ROCm tests (#44969)
Signed-off-by: Andreas Karatzas <akaratza@amd.com> Signed-off-by: Andreas Karatzas <Andreas.Karatzas@amd.com> Co-authored-by: OpenAI Codex <codex@openai.com>
Andreas Karatzas committed
Aug 18, 2026 at 17:45 UTC
8f4a7f45c53ab52b17023d3ca804e477daa36a23
11 files changed
+177
-1
.buildkite/test_areas/engine.yaml
+9
-1
@@ -133,7 +133,7 @@ steps:
133
depends_on:
134
- image-build-amd
135
136
-- label: ":nvidia: (H100) V1 E2E"
136
+- label: ":nvidia: (H100) V1 E2E Hybrid Chunked Prefill"
137
key: v1-e2e-4xh100
138
timeout_in_minutes: 35
139
device: h100
@@ -145,3 +145,11 @@ steps:
145
- tests/v1/e2e/test_hybrid_chunked_prefill.py
146
commands:
147
- pytest -v -s v1/e2e/test_hybrid_chunked_prefill.py
148
+ mirror:
149
+ amd:
150
+ label: ":amd: (MI300) V1 E2E Hybrid Chunked Prefill"
151
+ dind: false
152
+ device: mi300_4
153
+ timeout_in_minutes: 35
154
+ depends_on:
155
+ - image-build-amd
.buildkite/test_areas/entrypoints.yaml
+24
@@ -14,6 +14,14 @@ steps:
14
commands:
15
- pytest -v -s entrypoints/unit_tests
16
- pytest -v -s entrypoints/weight_transfer
17
+ mirror:
18
+ amd:
19
+ label: ":amd: (MI355) Entrypoints Unit"
20
+ dind: false
21
+ device: mi355_1
22
+ timeout_in_minutes: 35
23
+ depends_on:
24
+ - image-build-amd
25
26
- label: ":nvidia: (H200) Entrypoints Integration (LLM)"
27
device: h200_35gb
@@ -144,6 +152,14 @@ steps:
152
- tests/entrypoints/openai/responses
153
commands:
154
- pytest -v -s entrypoints/openai/responses
155
+ mirror:
156
+ amd:
157
+ label: ":amd: (MI355) Entrypoints Integration (Responses API)"
158
+ dind: false
159
+ device: mi355_1
160
+ timeout_in_minutes: 50
161
+ depends_on:
162
+ - image-build-amd
163
164
- label: ":nvidia: (H200) Entrypoints Integration (Speech to Text)"
165
device: h200_35gb
@@ -170,6 +186,14 @@ steps:
186
commands:
187
- export VLLM_WORKER_MULTIPROC_METHOD=spawn
188
- pytest -v -s entrypoints/multimodal
189
+ mirror:
190
+ amd:
191
+ label: ":amd: (MI355) Entrypoints Integration (Multimodal)"
192
+ dind: false
193
+ device: mi355_1
194
+ timeout_in_minutes: 55
195
+ depends_on:
196
+ - image-build-amd
197
198
- label: ":nvidia: (H200) Entrypoints Integration (Pooling)"
199
device: h200_35gb
.buildkite/test_areas/lora.yaml
+22
@@ -50,3 +50,25 @@ steps:
50
- pytest -v -s -x lora/test_gptoss_tp.py
51
- pytest -v -s -x lora/test_qwen35_densemodel_lora.py
52
- pytest -v -s -x lora/test_gemma4_tp.py
53
+ mirror:
54
+ amd:
55
+ label: ":amd: (MI300) LoRA TP (Distributed)"
56
+ dind: false
57
+ device: mi300_4
58
+ working_dir: "/vllm-workspace/tests"
59
+ timeout_in_minutes: 55
60
+ depends_on:
61
+ - image-build-amd
62
+ source_file_dependencies:
63
+ - vllm/lora
64
+ - vllm/model_executor/layers/fused_moe/
65
+ - vllm/platforms/rocm.py
66
+ - tests/lora
67
+ commands:
68
+ - pytest -v -s -x lora/test_chatglm3_tp.py
69
+ - pytest -v -s -x lora/test_llama_tp.py
70
+ - pytest -v -s -x lora/test_qwen3_with_multi_loras.py
71
+ - pytest -v -s -x lora/test_olmoe_tp.py
72
+ - pytest -v -s -x lora/test_gptoss_tp.py
73
+ - pytest -v -s -x lora/test_qwen35_densemodel_lora.py
74
+ - pytest -v -s -x lora/test_gemma4_tp.py
.buildkite/test_areas/misc.yaml
+8
@@ -317,6 +317,14 @@ steps:
317
- pytest -v -s detokenizer
318
- pytest -v -s -m 'not cpu_test' multimodal
319
- pytest -v -s utils_
320
+ mirror:
321
+ amd:
322
+ label: ":amd: (MI355) Async Engine, Inputs, Utils, Worker"
323
+ dind: false
324
+ device: mi355_1
325
+ timeout_in_minutes: 40
326
+ depends_on:
327
+ - image-build-amd
328
329
- label: ":computer: (CPU) Async Engine, Inputs, Utils, Worker, Config"
330
key: async-engine-inputs-utils-worker-config-cpu
.buildkite/test_areas/model_runner_v2.yaml
+21
@@ -81,6 +81,27 @@ steps:
81
- export NCCL_CUMEM_HOST_ENABLE=0
82
- TP_SIZE=1 DP_SIZE=2 pytest -v -s v1/distributed/test_async_llm_dp.py -k "not ray"
83
- TP_SIZE=1 DP_SIZE=2 pytest -v -s v1/distributed/test_eagle_dp.py
84
+ mirror:
85
+ amd:
86
+ label: ":amd: (MI300) Model Runner V2 Distributed"
87
+ dind: false
88
+ device: mi300_2
89
+ timeout_in_minutes: 45
90
+ depends_on:
91
+ - image-build-amd
92
+ source_file_dependencies:
93
+ - vllm/v1/worker/gpu/
94
+ - vllm/v1/worker/gpu_worker.py
95
+ - tests/basic_correctness/test_basic_correctness.py
96
+ - tests/v1/distributed/test_async_llm_dp.py
97
+ - tests/v1/distributed/test_eagle_dp.py
98
+ - vllm/platforms/rocm.py
99
+ commands:
100
+ - set -x
101
+ - export VLLM_USE_V2_MODEL_RUNNER=1
102
+ - TARGET_TEST_SUITE=MI300 pytest -v -s basic_correctness/test_basic_correctness.py -m 'distributed(num_gpus=2)' -k "not ray and not True"
103
+ - TP_SIZE=1 DP_SIZE=2 pytest -v -s v1/distributed/test_async_llm_dp.py -k "not ray"
104
+ - TP_SIZE=1 DP_SIZE=2 pytest -v -s v1/distributed/test_eagle_dp.py
105
106
- label: ":nvidia: (L4) Model Runner V2 Pipeline Parallelism"
107
key: model-runner-v2-pipeline-parallelism-4-gpus
.buildkite/test_areas/models_distributed.yaml
+27
@@ -22,3 +22,30 @@ steps:
22
- pytest models/multimodal/generation/test_phi4siglip.py -v -s -m 'distributed(num_gpus=2)'
23
- pytest models/multimodal -v -s -m 'distributed(num_gpus=2)' --ignore models/multimodal/generation/test_whisper.py --ignore models/multimodal/generation/test_phi4siglip.py
24
- VLLM_WORKER_MULTIPROC_METHOD=spawn pytest models/multimodal/generation/test_whisper.py -v -s -m 'distributed(num_gpus=2)'
25
+ mirror:
26
+ amd:
27
+ label: ":amd: (MI300) Distributed Models"
28
+ dind: false
29
+ device: mi300_2
30
+ timeout_in_minutes: 75
31
+ depends_on:
32
+ - image-build-amd
33
+ source_file_dependencies:
34
+ - vllm/model_executor/model_loader/sharded_state_loader.py
35
+ - vllm/model_executor/models/
36
+ - vllm/model_executor/layers/
37
+ - vllm/v1/attention/backends/
38
+ - vllm/v1/attention/selector.py
39
+ - vllm/_aiter_ops.py
40
+ - vllm/platforms/rocm.py
41
+ - tests/basic_correctness/
42
+ - tests/model_executor/model_loader/test_sharded_state_loader.py
43
+ - tests/models/
44
+ commands:
45
+ - TARGET_TEST_SUITE=MI300 pytest basic_correctness/ -v -s -m 'distributed(num_gpus=2)'
46
+ - HIP_VISIBLE_DEVICES=0,1 pytest -v -s model_executor/model_loader/test_sharded_state_loader.py -m '(not slow_test)'
47
+ - pytest models/transformers/test_backend.py -v -s -m 'distributed(num_gpus=2)'
48
+ - pytest models/language -v -s -m 'distributed(num_gpus=2)'
49
+ - pytest models/multimodal -v -s -m 'distributed(num_gpus=2)' --ignore models/multimodal/generation/test_whisper.py --ignore models/multimodal/generation/test_phi4siglip.py
50
+ - pytest models/multimodal/generation/test_phi4siglip.py -v -s -m 'distributed(num_gpus=2)'
51
+ - VLLM_WORKER_MULTIPROC_METHOD=spawn pytest models/multimodal/generation/test_whisper.py -v -s -m 'distributed(num_gpus=2)'
.buildkite/test_areas/models_language.yaml
+12
@@ -114,6 +114,18 @@ steps:
114
- MAMBA_FORCE_BUILD=TRUE uv pip install --system --no-build-isolation 'git+https://github.com/state-spaces/mamba@v2.3.0'
115
- CAUSAL_CONV1D_FORCE_BUILD=TRUE uv pip install --system --no-build-isolation 'git+https://github.com/Dao-AILab/causal-conv1d@v1.6.0'
116
- pytest -v -s models/language/generation -m '(not core_model) and (not hybrid_model)'
117
+ mirror:
118
+ amd:
119
+ label: ":amd: (MI355) Language Models (Extended Generation)"
120
+ dind: false
121
+ device: mi355_1
122
+ timeout_in_minutes: 70
123
+ depends_on:
124
+ - image-build-amd
125
+ commands:
126
+ - MAMBA_FORCE_BUILD=TRUE uv pip install --system --no-build-isolation 'git+https://github.com/AndreasKaratzas/mamba@fix-rocm-7.0-warp-size-constexpr'
127
+ - CAUSAL_CONV1D_FORCE_BUILD=TRUE uv pip install --system --no-build-isolation 'git+https://github.com/Dao-AILab/causal-conv1d@v1.6.0'
128
+ - pytest -v -s models/language/generation -m '(not core_model) and (not hybrid_model)'
129
130
- label: ":nvidia: (H200) Language Models (PPL)"
131
key: language-models-test-ppl
.buildkite/test_areas/models_multimodal.yaml
+24
@@ -110,6 +110,14 @@ steps:
110
- tests/models/registry.py
111
commands:
112
- pytest -v -s models/multimodal/processing/test_tensor_schema.py
113
+ mirror:
114
+ amd:
115
+ label: ":amd: (MI355) Multimodal Processor"
116
+ dind: false
117
+ device: mi355_1
118
+ timeout_in_minutes: 115
119
+ depends_on:
120
+ - image-build-amd
121
122
- label: ":nvidia: (H200) Multimodal Accuracy Eval (Small Models)"
123
device: h200_35gb
@@ -187,6 +195,14 @@ steps:
195
- tests/models/multimodal/generation
196
commands:
197
- pytest -v -s models/multimodal/generation/test_common.py -m 'split(group=0) and not core_model' --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
198
+ mirror:
199
+ amd:
200
+ label: ":amd: (MI355) Multimodal Models (Extended Generation 2) Shard %N"
201
+ dind: false
202
+ device: mi355_1
203
+ timeout_in_minutes: 100
204
+ depends_on:
205
+ - image-build-amd
206
207
- label: ":nvidia: (H200) Multimodal Models (Extended Generation 3)"
208
device: h200_35gb
@@ -198,6 +214,14 @@ steps:
214
- tests/models/multimodal/generation
215
commands:
216
- pytest -v -s models/multimodal/generation/test_common.py -m 'split(group=1) and not core_model'
217
+ mirror:
218
+ amd:
219
+ label: ":amd: (MI355) Multimodal Models (Extended Generation 3)"
220
+ dind: false
221
+ device: mi355_1
222
+ timeout_in_minutes: 90
223
+ depends_on:
224
+ - image-build-amd
225
226
- label: ":nvidia: (H200) Multimodal Models (Extended Pooling)"
227
key: multi-modal-models-extended-pooling
.buildkite/test_areas/plugins.yaml
+11
@@ -49,6 +49,17 @@ steps:
49
- pytest -v -s plugins_tests/test_oot_registration_online.py # it needs a clean process
50
- pytest -v -s plugins_tests/test_oot_registration_offline.py # it needs a clean process
51
- pytest -v -s plugins_tests/lora_resolvers # unit tests for in-tree lora resolver plugins
52
+ mirror:
53
+ amd:
54
+ label: ":amd: (MI250) Plugin Integration"
55
+ device: mi250_2
56
+ timeout_in_minutes: 50
57
+ depends_on:
58
+ - image-build-amd
59
+ source_file_dependencies:
60
+ - vllm/plugins/
61
+ - tests/plugins/
62
+ - vllm/platforms/rocm.py
63
64
65
- label: ":nvidia: (H200) GGUF Plugin"
.buildkite/test_areas/pytorch.yaml
+11
@@ -42,6 +42,17 @@ steps:
42
# However, find does not normally propagate error codes, so we combine it with xargs
43
# (using -0 for proper path handling)
44
- "find compile/ -maxdepth 1 -name 'test_*.py' -print0 | xargs -0 -n1 -I{} pytest -s -v '{}'"
45
+ mirror:
46
+ amd:
47
+ label: ":amd: (MI300) PyTorch Compilation"
48
+ dind: false
49
+ device: mi300_1
50
+ timeout_in_minutes: 90
51
+ depends_on:
52
+ - image-build-amd
53
+ source_file_dependencies:
54
+ - csrc/
55
+ - vllm/platforms/rocm.py
56
57
- label: ":nvidia: (H200) PyTorch Compilation H100 Cases"
58
key: pytorch-compilation-unit-tests-h100
.buildkite/test_areas/spec_decode.yaml
+8
@@ -223,6 +223,14 @@ steps:
223
commands:
224
- export VLLM_ALLOW_INSECURE_SERIALIZATION=1
225
- pytest -v -s v1/spec_decode/test_speculators_correctness.py -m slow_test
226
+ mirror:
227
+ amd:
228
+ label: ":amd: (MI300) Speculators Correctness Nightly"
229
+ dind: false
230
+ device: mi300_1
231
+ timeout_in_minutes: 40
232
+ depends_on:
233
+ - image-build-amd
234
235
- label: ":nvidia: (B200) Spec Decode DeepSeek MTP Parallel Load"
236
key: spec-decode-deepseek-mtp-parallel-load-2xb200-2xmi300