@tduncan / Nvidia-Optimization-Stack-VLLM / commits / 8d6b18329a

[CI] Standardize test job labels by device (#52659)

Signed-off-by: Kevin Luu <51931015+khluu@users.noreply.github.com> Co-authored-by: OpenAI Codex <codex@openai.com>

Kevin H. Luu committed Aug 18, 2026 at 12:50 UTC 8d6b18329a303c3d4a8f1a7b515cd69a4ef64800
35 files changed +268 -210
.buildkite/test_areas/attention.yaml
+3 -2
@@ -2,7 +2,7 @@ group: Attention
2 depends_on:
3 - image-build
4 steps:
5 -- label: V1 attention (H100-MI300)
5 +- label: ":nvidia: (H100) V1 Attention Shard %N"
6 key: v1-attention-h100-mi300
7 timeout_in_minutes: 85
8 device: h100
@@ -16,6 +16,7 @@ steps:
16 parallelism: 2
17 mirror:
18 amd:
19 + label: ":amd: (MI300) V1 Attention Shard %N"
20 dind: false
21 device: mi300_1
22 timeout_in_minutes: 125
@@ -30,7 +31,7 @@ steps:
31 - vllm/envs.py
32 - vllm/platforms/rocm.py
33
33 -- label: V1 attention (B200)
34 +- label: ":nvidia: (B200) V1 Attention Shard %N"
35 key: v1-attention-b200
36 timeout_in_minutes: 80
37 device: b200-k8s
.buildkite/test_areas/basic_correctness.yaml
+2 -1
@@ -2,7 +2,7 @@ group: Basic Correctness
2 depends_on:
3 - image-build
4 steps:
5 -- label: Basic Correctness
5 +- label: ":nvidia: (H200) Basic Correctness"
6 key: basic-correctness
7 timeout_in_minutes: 68
8 device: h200_18gb
@@ -19,6 +19,7 @@ steps:
19 - pytest -v -s basic_correctness/test_cpu_offload.py
20 mirror:
21 amd:
22 + label: ":amd: (MI300) Basic Correctness"
23 dind: false
24 device: mi300_1
25 timeout_in_minutes: 60
.buildkite/test_areas/benchmarks.yaml
+3 -2
@@ -2,7 +2,7 @@ group: Benchmarks
2 depends_on:
3 - image-build
4 steps:
5 -- label: Benchmarks CLI Test
5 +- label: ":nvidia: (H200) Benchmarks CLI"
6 key: benchmarks-cli-test
7 timeout_in_minutes: 45
8 device: h200_18gb
@@ -14,13 +14,14 @@ steps:
14 - pytest -v -s benchmarks/
15 mirror:
16 amd:
17 + label: ":amd: (MI300) Benchmarks CLI"
18 dind: false
19 device: mi300_1
20 timeout_in_minutes: 40
21 depends_on:
22 - image-build-amd
23
23 -- label: Attention Benchmarks Smoke Test (B200)
24 +- label: ":nvidia: (B200) Attention Benchmark Smoke"
25 key: attention-benchmarks-smoke-test-b200
26 device: b200-k8s
27 num_gpus: 2
.buildkite/test_areas/compile.yaml
+11 -11
@@ -2,7 +2,7 @@ group: Compile
2 depends_on:
3 - image-build
4 steps:
5 -- label: Sequence Parallel Correctness Tests (2xB200)
5 +- label: ":nvidia: (B200) Sequence Parallel Correctness"
6 key: sequence-parallel-correctness-tests-2xb200
7 timeout_in_minutes: 45
8 working_dir: "/vllm-workspace/"
@@ -13,7 +13,7 @@ steps:
13 - export VLLM_TEST_CLEAN_GPU_MEMORY=1
14 - pytest -v -s tests/compile/correctness_e2e/test_sequence_parallel.py
15
16 -- label: AsyncTP Correctness Tests (2xB200)
16 +- label: ":nvidia: (B200) AsyncTP Correctness"
17 key: asynctp-correctness-tests-b200
18 timeout_in_minutes: 30
19 working_dir: "/vllm-workspace/"
@@ -24,7 +24,7 @@ steps:
24 - export VLLM_TEST_CLEAN_GPU_MEMORY=1
25 - pytest -v -s tests/compile/correctness_e2e/test_async_tp.py
26
27 -- label: Distributed Compile Unit Tests (2xH100)
27 +- label: ":nvidia: (H100) Distributed Compile"
28 key: distributed-compile-unit-tests-2xh100
29 timeout_in_minutes: 45
30 working_dir: "/vllm-workspace/"
@@ -38,7 +38,7 @@ steps:
38 - export VLLM_TEST_CLEAN_GPU_MEMORY=1
39 - pytest -s -v tests/compile/passes/distributed
40
41 -- label: Fusion and Compile Unit Tests (2xB200)
41 +- label: ":nvidia: (B200) Fusion and Compile"
42 key: fusion-and-compile-unit-tests-2xb200
43 timeout_in_minutes: 30
44 working_dir: "/vllm-workspace/"
@@ -65,7 +65,7 @@ steps:
65 # this runner has 2 GPUs available even though num_devices=2 is not set
66 - pytest -v -s tests/compile/passes/distributed/test_fusion_all_reduce.py
67
68 -- label: Fusion E2E Quick (H100)
68 +- label: ":nvidia: (H100) Fusion E2E Quick"
69 key: fusion-e2e-quick-h100
70 timeout_in_minutes: 25
71 working_dir: "/vllm-workspace/"
@@ -84,7 +84,7 @@ steps:
84 # Qwen/Deepseek requires +quant_fp8 as -quant_fp8 rms+quant fusion is not supported
85 - pytest -v -s tests/compile/fusions_e2e/test_tp1_quant.py -k "inductor_partition and not +rms_norm and +quant_fp8 and (qwen3 or deepseek)"
86
87 -- label: Fusion E2E Config Sweep (H100)
87 +- label: ":nvidia: (H100) Fusion E2E Config Sweep"
88 key: fusion-e2e-config-sweep-h100
89 timeout_in_minutes: 25
90 working_dir: "/vllm-workspace/"
@@ -104,7 +104,7 @@ steps:
104 # Run just llama3 (fp8) for all config combinations
105 - pytest -v -s tests/compile/fusions_e2e/test_tp1_quant.py -k "llama-3"
106
107 -- label: Fusion E2E Config Sweep (B200)
107 +- label: ":nvidia: (B200) Fusion E2E Config Sweep"
108 key: fusion-e2e-config-sweep-b200
109 timeout_in_minutes: 30
110 working_dir: "/vllm-workspace/"
@@ -118,7 +118,7 @@ steps:
118 # Run just llama3 (fp8 & fp4) for all config combinations (only inductor partition)
119 - pytest -v -s tests/compile/fusions_e2e/test_tp1_quant.py -k "inductor_partition and (FLASHINFER and not +rms_norm and (not +quant_fp8 or +quant_fp8 and (qwen3 or deepseek)) or llama-3)"
120
121 -- label: Fusion E2E TP2 Quick (H100)
121 +- label: ":nvidia: (H100) Fusion E2E TP2 Quick"
122 key: fusion-e2e-tp2-quick-h100
123 timeout_in_minutes: 35
124 working_dir: "/vllm-workspace/"
@@ -136,7 +136,7 @@ steps:
136 - pytest -v -s tests/compile/fusions_e2e/test_tp2_ar_rms.py -k "inductor_partition and not +rms_norm and (not +quant_fp8 or +quant_fp8 and (qwen3 or deepseek))"
137 - pytest -v -s tests/compile/fusions_e2e/test_tp2_async_tp.py -k "inductor_partition and not +rms_norm and (not +quant_fp8 or +quant_fp8 and (qwen3 or deepseek))"
138
139 -- label: Fusion E2E TP2 AR-RMS Config Sweep (H100)
139 +- label: ":nvidia: (H100) Fusion E2E TP2 AR-RMS Config Sweep"
140 key: fusion-e2e-tp2-ar-rms-config-sweep-h100
141 timeout_in_minutes: 30
142 working_dir: "/vllm-workspace/"
@@ -156,7 +156,7 @@ steps:
156 # Run just llama3 (fp8 & bf16) for all config combinations
157 - pytest -v -s tests/compile/fusions_e2e/test_tp2_ar_rms.py -k "llama-3"
158
159 -- label: Fusion E2E TP2 AsyncTP Config Sweep (H100)
159 +- label: ":nvidia: (H100) Fusion E2E TP2 AsyncTP Config Sweep"
160 key: fusion-e2e-tp2-asynctp-config-sweep-h100
161 timeout_in_minutes: 40
162 working_dir: "/vllm-workspace/"
@@ -176,7 +176,7 @@ steps:
176 # Run just llama3 (fp8 & bf16) for all config combinations
177 - pytest -v -s tests/compile/fusions_e2e/test_tp2_async_tp.py -k "llama-3"
178
179 -- label: Fusion E2E TP2 (B200)
179 +- label: ":nvidia: (B200) Fusion E2E TP2"
180 key: fusion-e2e-tp2-b200
181 timeout_in_minutes: 45
182 working_dir: "/vllm-workspace/"
.buildkite/test_areas/cuda.yaml
+2 -2
@@ -2,7 +2,7 @@ group: CUDA
2 depends_on:
3 - image-build
4 steps:
5 -- label: Platform Tests
5 +- label: ":nvidia: (H200) CUDA Platform"
6 key: platform-tests
7 timeout_in_minutes: 20
8 device: h200_18gb
@@ -18,7 +18,7 @@ steps:
18 - pytest -v -s cuda/test_platform_no_cuda_init.py
19 - pytest -v -s cuda/test_cuda_compatibility_path.py
20
21 -- label: Cudagraph
21 +- label: ":nvidia: (H200) CUDAGraph"
22 device: h200_35gb
23 key: cudagraph
24 timeout_in_minutes: 30
.buildkite/test_areas/disaggregated.yaml
+18 -13
@@ -2,7 +2,7 @@ group: Disaggregated
2 depends_on:
3 - image-build
4 steps:
5 -- label: Distributed NixlConnector PD accuracy (4 GPUs)
5 +- label: ":nvidia: (L4) Distributed NixlConnector PD accuracy"
6 key: distributed-nixlconnector-pd-accuracy-4-gpus
7 timeout_in_minutes: 55
8 working_dir: "/vllm-workspace/tests"
@@ -15,6 +15,7 @@ steps:
15 - bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh
16 mirror:
17 amd:
18 + label: ":amd: (MI300) Distributed NixlConnector PD accuracy"
19 dind: false
20 device: mi300_4
21 timeout_in_minutes: 60
@@ -28,7 +29,7 @@ steps:
29 - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt
30 - ATTENTION_BACKEND=TRITON_ATTN bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh
31
31 -- label: Distributed FlashInfer NixlConnector PD accuracy (4 GPUs)
32 +- label: ":nvidia: (L4) Distributed FlashInfer NixlConnector PD accuracy"
33 key: distributed-flashinfer-nixlconnector-pd-accuracy-4-gpus
34 timeout_in_minutes: 55
35 working_dir: "/vllm-workspace/tests"
@@ -40,7 +41,7 @@ steps:
41 - bash /vllm-workspace/.buildkite/scripts/install-kv-connectors.sh
42 - FLASHINFER=1 bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh
43
43 -- label: Push NixlConnector PP prefill PD accuracy (4 GPUs)
44 +- label: ":nvidia: (L4) Push NixlConnector PP prefill PD accuracy"
45 key: push-nixlconnector-pp-prefill-pd-accuracy-4-gpus
46 timeout_in_minutes: 30
47 working_dir: "/vllm-workspace/tests"
@@ -53,7 +54,7 @@ steps:
54 - bash /vllm-workspace/.buildkite/scripts/install-kv-connectors.sh
55 - bash v1/kv_connector/nixl_push_integration/config_sweep_accuracy_test.sh
56
56 -- label: DP EP Distributed NixlConnector PD accuracy tests (4 GPUs)
57 +- label: ":nvidia: (L4) DP EP Distributed NixlConnector PD accuracy"
58 key: dp-ep-distributed-nixlconnector-pd-accuracy-tests-4-gpus
59 timeout_in_minutes: 30
60 working_dir: "/vllm-workspace/tests"
@@ -66,6 +67,7 @@ steps:
67 - DP_EP=1 bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh
68 mirror:
69 amd:
70 + label: ":amd: (MI300) DP EP Distributed NixlConnector PD accuracy"
71 dind: false
72 device: mi300_4
73 timeout_in_minutes: 40
@@ -79,7 +81,7 @@ steps:
81 - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt
82 - DP_EP=1 ATTENTION_BACKEND=TRITON_ATTN bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh
83
82 -- label: CrossLayer KV layout Distributed NixlConnector PD accuracy tests (4 GPUs)
84 +- label: ":nvidia: (L4) CrossLayer KV layout Distributed NixlConnector PD accuracy"
85 key: crosslayer-kv-layout-distributed-nixlconnector-pd-accuracy-tests-4-gpus
86 timeout_in_minutes: 55
87 working_dir: "/vllm-workspace/tests"
@@ -92,6 +94,7 @@ steps:
94 - CROSS_LAYERS_BLOCKS=True bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh
95 mirror:
96 amd:
97 + label: ":amd: (MI300) CrossLayer KV layout Distributed NixlConnector PD accuracy"
98 dind: false
99 device: mi300_4
100 timeout_in_minutes: 60
@@ -105,7 +108,7 @@ steps:
108 - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt
109 - CROSS_LAYERS_BLOCKS=True ATTENTION_BACKEND=TRITON_ATTN bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh
110
108 -- label: Hybrid SSM NixlConnector PD accuracy tests (4 GPUs)
111 +- label: ":nvidia: (L4) Hybrid SSM NixlConnector PD accuracy"
112 key: hybrid-ssm-nixlconnector-pd-accuracy-tests-4-gpus
113 timeout_in_minutes: 60
114 working_dir: "/vllm-workspace/tests"
@@ -118,6 +121,7 @@ steps:
121 - HYBRID_SSM=1 bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh
122 mirror:
123 amd:
124 + label: ":amd: (MI300) Hybrid SSM NixlConnector PD accuracy"
125 dind: false
126 device: mi300_4
127 timeout_in_minutes: 55
@@ -131,7 +135,7 @@ steps:
135 - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt
136 - HYBRID_SSM=1 ATTENTION_BACKEND=TRITON_ATTN bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh
137
134 -- label: NixlConnector PD edge case test (2 GPUs)
138 +- label: ":nvidia: (L4) NixlConnector PD edge case"
139 key: nixlconnector-pd-edge-cases-2-gpus
140 timeout_in_minutes: 40
141 working_dir: "/vllm-workspace/tests"
@@ -147,7 +151,7 @@ steps:
151 - bash /vllm-workspace/.buildkite/scripts/install-kv-connectors.sh
152 - bash v1/kv_connector/nixl_integration/run_edge_case_test.sh
153
150 -- label: Hybrid SSM NixlConnector PD prefix cache test (2 GPUs)
154 +- label: ":nvidia: (L4) Hybrid SSM NixlConnector PD prefix cache"
155 key: hybrid-ssm-nixlconnector-pd-prefix-cache-2-gpus
156 timeout_in_minutes: 25
157 working_dir: "/vllm-workspace/tests"
@@ -161,7 +165,7 @@ steps:
165 - bash /vllm-workspace/.buildkite/scripts/install-kv-connectors.sh
166 - bash v1/kv_connector/nixl_integration/run_mamba_prefix_cache_test.sh
167
164 -- label: MultiConnector (Nixl+Offloading) PD accuracy (2 GPUs)
168 +- label: ":nvidia: (L4) MultiConnector (Nixl+Offloading) PD accuracy"
169 key: multiconnector-nixl-offloading-pd-accuracy-2-gpus
170 timeout_in_minutes: 40
171 working_dir: "/vllm-workspace/tests"
@@ -176,7 +180,7 @@ steps:
180 - bash /vllm-workspace/.buildkite/scripts/install-kv-connectors.sh
181 - bash v1/kv_connector/nixl_integration/run_multi_connector_accuracy_test.sh
182
179 -- label: NixlConnector PD + Spec Decode acceptance (2 GPUs)
183 +- label: ":nvidia: (A100) NixlConnector PD + Spec Decode acceptance"
184 key: nixlconnector-pd-spec-decode-acceptance-2-gpus
185 timeout_in_minutes: 45
186 device: a100
@@ -191,6 +195,7 @@ steps:
195 - bash v1/kv_connector/nixl_integration/config_sweep_spec_decode_test.sh
196 mirror:
197 amd:
198 + label: ":amd: (MI300) NixlConnector PD + Spec Decode acceptance"
199 dind: false
200 device: mi300_2
201 timeout_in_minutes: 45
@@ -205,7 +210,7 @@ steps:
210 - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt
211 - KV_CACHE_MEMORY_BYTES=8G ATTENTION_BACKEND=TRITON_ATTN bash v1/kv_connector/nixl_integration/config_sweep_spec_decode_test.sh
212
208 -- label: MultiConnector (Nixl+Offloading) PD edge cases (2 GPUs)
213 +- label: ":nvidia: (L4) MultiConnector (Nixl+Offloading) PD edge cases"
214 key: multiconnector-nixl-offloading-pd-edge-cases-2-gpus
215 timeout_in_minutes: 25
216 working_dir: "/vllm-workspace/tests"
@@ -221,7 +226,7 @@ steps:
226 - bash v1/kv_connector/nixl_integration/run_multi_connector_edge_case_test.sh
227
228 # P TP 4 - D DPEP 4 test case for DSv4-Flash
224 -- label: DSv4-Flash Disaggregated DP EP
229 +- label: ":nvidia: (H200) DSv4-Flash Disaggregated DP EP"
230 key: dsv4-flash-disaggregated
231 timeout_in_minutes: 60
232 device: h200
@@ -242,7 +247,7 @@ steps:
247 - bash /vllm-workspace/.buildkite/scripts/install-kv-connectors.sh
248 - bash v1/kv_connector/nixl_integration/run_accuracy_test.sh
249
245 -- label: Kimi-Linear-48B-A3B Disaggregated DP EP
250 +- label: ":nvidia: (H200) Kimi-Linear-48B-A3B Disaggregated DP EP"
251 key: kimi-linear-disaggregated
252 timeout_in_minutes: 60
253 device: h200
.buildkite/test_areas/disaggregated_mooncake.yaml
+1 -1
@@ -2,7 +2,7 @@ group: Disaggregated Mooncake
2 depends_on:
3 - image-build
4 steps:
5 -- label: Distributed MooncakeConnector PD accuracy (4 GPUs)
5 +- label: ":nvidia: (B200) Distributed MooncakeConnector PD accuracy"
6 key: distributed-mooncakeconnector-pd-accuracy-4-gpus
7 timeout_in_minutes: 10
8 working_dir: "/vllm-workspace/tests"
.buildkite/test_areas/distributed.yaml
+15 -14
@@ -2,7 +2,7 @@ group: Distributed
2 depends_on:
3 - image-build
4 steps:
5 -- label: Distributed Comm Ops
5 +- label: ":nvidia: (L4) Distributed Comm Ops"
6 key: distributed-comm-ops
7 timeout_in_minutes: 25
8 working_dir: "/vllm-workspace/tests"
@@ -17,7 +17,7 @@ steps:
17 - pytest -v -s distributed/test_shm_buffer.py
18 - pytest -v -s distributed/test_shm_storage.py
19
20 -- label: Distributed DP Tests (2 GPUs)
20 +- label: ":nvidia: (L4) Distributed DP Basic"
21 key: distributed-dp-tests-2-gpus
22 timeout_in_minutes: 35
23 working_dir: "/vllm-workspace/tests"
@@ -41,6 +41,7 @@ steps:
41 - DP_SIZE=2 pytest -v -s entrypoints/openai/test_multi_api_servers.py
42 mirror:
43 amd:
44 + label: ":amd: (MI300) Distributed DP Basic"
45 dind: false
46 device: mi300_2
47 timeout_in_minutes: 45
@@ -58,7 +59,7 @@ steps:
59 - tests/entrypoints/openai/test_multi_api_servers.py
60 - vllm/platforms/rocm.py
61
61 -- label: Distributed Compile + RPC Tests (2 GPUs)
62 +- label: ":nvidia: (L4) Distributed Compile + RPC"
63 key: distributed-compile-rpc-tests-2-gpus
64 timeout_in_minutes: 65
65 working_dir: "/vllm-workspace/tests"
@@ -80,7 +81,7 @@ steps:
81 - pytest -v -s entrypoints/llm/test_collective_rpc.py
82 - pytest -v -s ./compile/test_wrapper.py
83
83 -- label: Distributed Torchrun + Shutdown Tests (2 GPUs)
84 +- label: ":nvidia: (L4) Distributed Torchrun + Shutdown"
85 key: distributed-torchrun-shutdown-tests-2-gpus
86 timeout_in_minutes: 30
87 working_dir: "/vllm-workspace/tests"
@@ -104,7 +105,7 @@ steps:
105 - CUDA_VISIBLE_DEVICES=0,1 pytest -v -s v1/shutdown
106 - pytest -v -s v1/worker/test_worker_memory_snapshot.py
107
107 -- label: Distributed Torchrun + Examples (4 GPUs)
108 +- label: ":nvidia: (L4) Distributed Torchrun + Examples"
109 key: distributed-torchrun-examples-4-gpus
110 timeout_in_minutes: 30
111 working_dir: "/vllm-workspace"
@@ -137,7 +138,7 @@ steps:
138 - VLLM_ALLOW_INSECURE_SERIALIZATION=1 python3 examples/rl/rlhf_http_nccl.py
139 - VLLM_ALLOW_INSECURE_SERIALIZATION=1 python3 examples/rl/rlhf_http_ipc.py
140
140 -- label: Distributed DP Tests (4 GPUs)
141 +- label: ":nvidia: (L4) Distributed DP Extended"
142 key: distributed-dp-tests-4-gpus
143 timeout_in_minutes: 45
144 working_dir: "/vllm-workspace/tests"
@@ -160,7 +161,7 @@ steps:
161 - pytest -v -s v1/engine/test_engine_core_client.py::test_kv_cache_events_dp
162 - pytest -v -s distributed/test_utils.py
163
163 -- label: Distributed Compile + Comm (4 GPUs)
164 +- label: ":nvidia: (L4) Distributed Compile + Comm"
165 key: distributed-compile-comm-4-gpus
166 timeout_in_minutes: 70
167 working_dir: "/vllm-workspace/tests"
@@ -181,7 +182,7 @@ steps:
182 # test multi-node TP with multiproc executor (simulated on single node)
183 - pytest -v -s distributed/test_multiproc_executor.py::test_multiproc_executor_multi_node
184
184 -- label: Distributed Tests (8xH100)
185 +- label: ":nvidia: (H100) Distributed DP + EP"
186 key: distributed-tests-8xh100
187 timeout_in_minutes: 20
188 device: h100
@@ -203,7 +204,7 @@ steps:
204 # test with torchrun tp=2 and dp=4 with ep
205 - torchrun --nproc-per-node=8 ../examples/features/torchrun/torchrun_dp_example_offline.py --tp-size=2 --pp-size=1 --dp-size=4 --enable-ep
206
206 -- label: Distributed Tests (4xA100)
207 +- label: ":nvidia: (A100) Distributed"
208 key: distributed-tests-4xa100
209 device: a100
210 optional: true
@@ -219,7 +220,7 @@ steps:
220 - TARGET_TEST_SUITE=A100 pytest basic_correctness/ -v -s -m 'distributed(num_gpus=2)'
221 - pytest -v -s -x lora/test_mixtral.py
222
222 -- label: Distributed Tests (2xH100-2xMI300)
223 +- label: ":nvidia: (H100) Distributed Features"
224 key: distributed-tests-2xh100-2xmi300
225 timeout_in_minutes: 30
226 device: h100
@@ -234,7 +235,7 @@ steps:
235 - VLLM_ALLOW_INSECURE_SERIALIZATION=1 pytest -v -s tests/distributed/test_weight_transfer.py
236 - pytest -v -s tests/distributed/test_packed_tensor.py
237
237 -- label: Distributed Tests (2xB200)
238 +- label: ":nvidia: (B200) Distributed"
239 key: distributed-tests-2xb200
240 device: b200-k8s
241 optional: true
@@ -249,7 +250,7 @@ steps:
250
251
252
252 -- label: 2 Node Test (4 GPUs)
253 +- label: ":nvidia: (L4) Distributed 2-Node"
254 key: 2-node-test-4-gpus
255 timeout_in_minutes: 30
256 working_dir: "/vllm-workspace/tests"
@@ -268,7 +269,7 @@ steps:
269 commands:
270 - ./.buildkite/scripts/run-multi-node-test.sh /vllm-workspace/tests 2 2 $IMAGE_TAG "VLLM_TEST_SAME_HOST=0 torchrun --nnodes 2 --nproc-per-node=2 --rdzv_backend=c10d --rdzv_endpoint=192.168.10.10 distributed/test_same_node.py | grep 'Same node test passed' && NUM_NODES=2 torchrun --nnodes 2 --nproc-per-node=2 --rdzv_backend=c10d --rdzv_endpoint=192.168.10.10 distributed/test_node_count.py | grep 'Node count test passed' && python3 ../examples/features/data_parallel/data_parallel_offline.py -dp=2 -tp=1 --dp-num-nodes=2 --dp-node-rank=0 --dp-master-addr=192.168.10.10 --dp-master-port=12345 --enforce-eager --trust-remote-code && VLLM_MULTI_NODE=1 pytest -v -s distributed/test_multi_node_assignment.py && VLLM_MULTI_NODE=1 pytest -v -s distributed/test_pipeline_parallel.py" "VLLM_TEST_SAME_HOST=0 torchrun --nnodes 2 --nproc-per-node=2 --rdzv_backend=c10d --rdzv_endpoint=192.168.10.10 distributed/test_same_node.py | grep 'Same node test passed' && NUM_NODES=2 torchrun --nnodes 2 --nproc-per-node=2 --rdzv_backend=c10d --rdzv_endpoint=192.168.10.10 distributed/test_node_count.py | grep 'Node count test passed' && python3 ../examples/features/data_parallel/data_parallel_offline.py -dp=2 -tp=1 --dp-num-nodes=2 --dp-node-rank=1 --dp-master-addr=192.168.10.10 --dp-master-port=12345 --enforce-eager --trust-remote-code"
271
271 -- label: Pipeline + Context Parallelism (4 GPUs)
272 +- label: ":nvidia: (L4) Pipeline + Context Parallelism"
273 key: pipeline-context-parallelism-4-gpus
274 timeout_in_minutes: 55
275 working_dir: "/vllm-workspace/tests"
@@ -284,7 +285,7 @@ steps:
285 - pytest -v -s distributed/test_pp_cudagraph.py
286 - pytest -v -s distributed/test_pipeline_parallel.py
287
287 -- label: RayExecutorV2 (4 GPUs)
288 +- label: ":nvidia: (L4) RayExecutorV2"
289 key: rayexecutorv2-4-gpus
290 timeout_in_minutes: 45
291 working_dir: "/vllm-workspace/tests"
.buildkite/test_areas/docker.yaml
+1 -1
@@ -2,7 +2,7 @@ group: Docker
2 depends_on:
3 - image-build-cpu
4 steps:
5 -- label: Docker Build Metadata
5 +- label: ":computer: (CPU) Docker Build Metadata"
6 timeout_in_minutes: 20
7 device: cpu-small
8 source_file_dependencies:
.buildkite/test_areas/e2e_integration.yaml
+5 -5
@@ -2,7 +2,7 @@ group: E2E Integration
2 depends_on:
3 - image-build
4 steps:
5 -- label: DeepSeek V2-Lite Sync EPLB Accuracy (4xH100)
5 +- label: ":nvidia: (H100) DeepSeek V2-Lite Sync EPLB Accuracy"
6 key: deepseek-v2-lite-sync-eplb-accuracy-4xh100
7 timeout_in_minutes: 25
8 device: h100
@@ -12,7 +12,7 @@ steps:
12 commands:
13 - bash .buildkite/scripts/scheduled_integration_test/deepseek_v2_lite_ep_eplb.sh 0.25 200 8010
14
15 -- label: Qwen3-30B-A3B-FP8-block Sync EPLB Accuracy (4xH100)
15 +- label: ":nvidia: (H100) Qwen3-30B-A3B-FP8-block Sync EPLB Accuracy"
16 key: qwen3-30b-a3b-fp8-block-sync-eplb-accuracy-4xh100
17 timeout_in_minutes: 25
18 device: h100
@@ -22,7 +22,7 @@ steps:
22 commands:
23 - bash .buildkite/scripts/scheduled_integration_test/qwen30b_a3b_fp8_block_ep_eplb.sh 0.8 200 8020
24
25 -- label: Qwen3-30B-A3B-FP8-block Sync EPLB Accuracy (2xB200)
25 +- label: ":nvidia: (B200) Qwen3-30B-A3B-FP8-block Sync EPLB Accuracy"
26 key: qwen3-30b-a3b-fp8-block-sync-eplb-accuracy-2xb200
27 timeout_in_minutes: 20
28 device: b200-k8s
@@ -32,7 +32,7 @@ steps:
32 commands:
33 - bash .buildkite/scripts/scheduled_integration_test/qwen30b_a3b_fp8_block_ep_eplb.sh 0.8 200 8020 2 1
34
35 -- label: Qwen3-30B-A3B-FP8 DP4 Async EPLB Accuracy
35 +- label: ":nvidia: (H100) Qwen3-30B-A3B-FP8 DP4 Async EPLB Accuracy"
36 key: qwen3-30b-a3b-fp8-dp4-async-eplb-accuracy
37 timeout_in_minutes: 25
38 device: h100
@@ -42,7 +42,7 @@ steps:
42 commands:
43 - bash .buildkite/scripts/scheduled_integration_test/qwen30b_a3b_fp8_dp4_async_eplb.sh 0.8 200 8050
44
45 -- label: DeepSeek V2-Lite Prefetch Offload Accuracy (H100)
45 +- label: ":nvidia: (H100) DeepSeek V2-Lite Prefetch Offload Accuracy"
46 key: deepseek-v2-lite-prefetch-offload-accuracy-h100
47 timeout_in_minutes: 20
48 device: h100
.buildkite/test_areas/engine.yaml
+11 -6
@@ -2,7 +2,7 @@ group: Engine
2 depends_on:
3 - image-build
4 steps:
5 -- label: Engine
5 +- label: ":nvidia: (H200) Engine Core"
6 key: engine
7 timeout_in_minutes: 30
8 device: h200_18gb
@@ -29,13 +29,14 @@ steps:
29 - pytest -v -s engine test_sequence.py test_config.py test_logger.py test_vllm_port.py jit_monitor/test_hooks.py jit_monitor/test_hooks_gpu.py
30 mirror:
31 amd:
32 + label: ":amd: (MI300) Engine Core"
33 dind: false
34 device: mi300_1
35 timeout_in_minutes: 40
36 depends_on:
37 - image-build-amd
38
38 -- label: Engine (1 GPU)
39 +- label: ":nvidia: (L4) V1 Engine"
40 key: engine-1-gpu
41 timeout_in_minutes: 45
42 source_file_dependencies:
@@ -48,12 +49,13 @@ steps:
49 - pytest -v -s v1/test_tensor_ipc_queue.py
50 mirror:
51 amd:
52 + label: ":amd: (MI250) V1 Engine"
53 device: mi250_1
54 timeout_in_minutes: 45
55 depends_on:
56 - image-build-amd
57
56 -- label: e2e Scheduling (1 GPU)
58 +- label: ":nvidia: (H200) E2E Scheduling"
59 key: e2e-scheduling-1-gpu
60 timeout_in_minutes: 53
61 device: h200_18gb
@@ -64,12 +66,13 @@ steps:
66 - pytest -v -s v1/e2e/general/test_async_scheduling.py
67 mirror:
68 amd:
69 + label: ":amd: (MI250) E2E Scheduling"
70 device: mi250_1
71 timeout_in_minutes: 55
72 depends_on:
73 - image-build-amd
74
72 -- label: e2e Core (1 GPU)
75 +- label: ":nvidia: (H200) E2E Core"
76 device: h200_35gb
77 key: e2e-core-1-gpu
78 timeout_in_minutes: 40
@@ -80,6 +83,7 @@ steps:
83 - pytest -v -s v1/e2e/general --ignore v1/e2e/general/test_async_scheduling.py
84 mirror:
85 amd:
86 + label: ":amd: (MI250) E2E Core"
87 device: mi250_1
88 timeout_in_minutes: 50
89 depends_on:
@@ -89,7 +93,7 @@ steps:
93 - tests/v1/e2e/general/
94 - vllm/platforms/rocm.py
95
92 -- label: V1 e2e (2 GPUs)
96 +- label: ":nvidia: (L4) V1 E2E"
97 key: v1-e2e-2-gpus
98 timeout_in_minutes: 25 # TODO: Fix timeout after we have more confidence in the test stability
99 optional: true
@@ -122,13 +126,14 @@ steps:
126 v1/e2e/spec_decode/draft_model/test_draft_model.py::test_draft_model_engine_args_tensor_parallelism
127 mirror:
128 amd:
129 + label: ":amd: (MI300) V1 E2E"
130 dind: false
131 device: mi300_2
132 timeout_in_minutes: 30
133 depends_on:
134 - image-build-amd
135
131 -- label: V1 e2e (4xH100)
136 +- label: ":nvidia: (H100) V1 E2E"
137 key: v1-e2e-4xh100
138 timeout_in_minutes: 35
139 device: h100
.buildkite/test_areas/entrypoints.yaml
+17 -11
@@ -2,7 +2,7 @@ group: Entrypoints
2 depends_on:
3 - image-build
4 steps:
5 -- label: Entrypoints Unit Tests
5 +- label: ":nvidia: (H200) Entrypoints Unit"
6 device: h200_35gb
7 key: entrypoints-unit-tests
8 timeout_in_minutes: 25
@@ -15,7 +15,7 @@ steps:
15 - pytest -v -s entrypoints/unit_tests
16 - pytest -v -s entrypoints/weight_transfer
17
18 -- label: Entrypoints Integration (LLM)
18 +- label: ":nvidia: (H200) Entrypoints Integration (LLM)"
19 device: h200_35gb
20 key: entrypoints-integration-llm
21 timeout_in_minutes: 60
@@ -31,13 +31,14 @@ steps:
31 - pytest -v -s entrypoints/llm/offline_mode # Needs to avoid interference with other tests
32 mirror:
33 amd:
34 + label: ":amd: (MI300) Entrypoints Integration (LLM)"
35 dind: false
36 device: mi300_1
37 timeout_in_minutes: 55
38 depends_on:
39 - image-build-amd
40
40 -- label: Entrypoints Integration (API Server)
41 +- label: ":nvidia: (H200) Entrypoints Integration (API Server)"
42 key: entrypoints-integration-api-server
43 device: h200_35gb
44 timeout_in_minutes: 75
@@ -54,13 +55,14 @@ steps:
55 - pytest -v -s entrypoints/scale_out
56 mirror:
57 amd:
58 + label: ":amd: (MI300) Entrypoints Integration (API Server)"
59 dind: false
60 device: mi300_1
61 timeout_in_minutes: 65
62 depends_on:
63 - image-build-amd
64
63 -- label: Entrypoints Integration (API Server OpenAI - Part 1)
65 +- label: ":nvidia: (H200) Entrypoints Integration (API Server OpenAI - Part 1)"
66 device: h200_35gb
67 key: entrypoints-integration-api-server-openai-part-1
68 timeout_in_minutes: 68
@@ -75,13 +77,14 @@ steps:
77 - pytest -v -s entrypoints/openai --ignore=entrypoints/openai/completion --ignore=entrypoints/openai/chat_completion --ignore=entrypoints/openai/responses --ignore=entrypoints/openai/correctness
78 mirror:
79 amd:
80 + label: ":amd: (MI300) Entrypoints Integration (API Server OpenAI - Part 1)"
81 dind: false
82 device: mi300_1
83 timeout_in_minutes: 65
84 depends_on:
85 - image-build-amd
86
84 -- label: Entrypoints Integration (API Server OpenAI - Part 2)
87 +- label: ":nvidia: (H200) Entrypoints Integration (API Server OpenAI - Part 2)"
88 device: h200_35gb
89 key: entrypoints-integration-api-server-openai-part-2
90 timeout_in_minutes: 83
@@ -97,13 +100,14 @@ steps:
100 - pytest -v -s entrypoints/openai/completion --ignore=entrypoints/openai/completion/test_tensorizer_entrypoint.py
101 mirror:
102 amd:
103 + label: ":amd: (MI300) Entrypoints Integration (API Server OpenAI - Part 2)"
104 dind: false
105 device: mi300_1
106 timeout_in_minutes: 70
107 depends_on:
108 - image-build-amd
109
106 -- label: Entrypoints Integration (API Server Generate)
110 +- label: ":nvidia: (H200) Entrypoints Integration (API Server Generate)"
111 device: h200_35gb
112 key: entrypoints-integration-api-server-generate
113 timeout_in_minutes: 50
@@ -122,13 +126,14 @@ steps:
126 - pytest -v -s entrypoints/anthropic
127 mirror:
128 amd:
129 + label: ":amd: (MI300) Entrypoints Integration (API Server Generate)"
130 dind: false
131 device: mi300_1
132 timeout_in_minutes: 65
133 depends_on:
134 - image-build-amd
135
131 -- label: Entrypoints Integration (Responses API)
136 +- label: ":nvidia: (H200) Entrypoints Integration (Responses API)"
137 device: h200_35gb
138 key: entrypoints-integration-responses-api
139 timeout_in_minutes: 50
@@ -140,7 +145,7 @@ steps:
145 commands:
146 - pytest -v -s entrypoints/openai/responses
147
143 -- label: Entrypoints Integration (Speech to Text)
148 +- label: ":nvidia: (H200) Entrypoints Integration (Speech to Text)"
149 device: h200_35gb
150 key: entrypoints-integration-speech_to_text
151 timeout_in_minutes: 45
@@ -153,7 +158,7 @@ steps:
158 - export VLLM_WORKER_MULTIPROC_METHOD=spawn
159 - pytest -v -s entrypoints/speech_to_text
160
156 -- label: Entrypoints Integration (Multimodal)
161 +- label: ":nvidia: (H200) Entrypoints Integration (Multimodal)"
162 device: h200_35gb
163 key: entrypoints-integration-multimodal
164 timeout_in_minutes: 45
@@ -166,7 +171,7 @@ steps:
171 - export VLLM_WORKER_MULTIPROC_METHOD=spawn
172 - pytest -v -s entrypoints/multimodal
173
169 -- label: Entrypoints Integration (Pooling)
174 +- label: ":nvidia: (H200) Entrypoints Integration (Pooling)"
175 device: h200_35gb
176 key: entrypoints-integration-pooling
177 timeout_in_minutes: 75
@@ -179,7 +184,7 @@ steps:
184 - export VLLM_WORKER_MULTIPROC_METHOD=spawn
185 - pytest -v -s entrypoints/pooling
186
182 -- label: OpenAI API Correctness
187 +- label: ":nvidia: (H200) OpenAI API Correctness"
188 key: openai-api-correctness
189 timeout_in_minutes: 20
190 device: h200_18gb
@@ -190,6 +195,7 @@ steps:
195 - pytest -s entrypoints/openai/correctness/
196 mirror:
197 amd:
198 + label: ":amd: (MI300) OpenAI API Correctness"
199 dind: false
200 device: mi300_1
201 timeout_in_minutes: 30
.buildkite/test_areas/expert_parallelism.yaml
+4 -3
@@ -2,7 +2,7 @@ group: Expert Parallelism
2 depends_on:
3 - image-build
4 steps:
5 -- label: EPLB Algorithm
5 +- label: ":nvidia: (H200) EPLB Algorithm"
6 key: eplb-algorithm
7 timeout_in_minutes: 20
8 device: h200_18gb
@@ -16,6 +16,7 @@ steps:
16 - pytest -v -s distributed/test_eplb_utils.py
17 mirror:
18 amd:
19 + label: ":amd: (MI300) EPLB Algorithm"
20 dind: false
21 device: mi300_1
22 timeout_in_minutes: 30
@@ -27,7 +28,7 @@ steps:
28 - tests/distributed/test_eplb_utils.py
29 - vllm/platforms/rocm.py
30
30 -- label: EPLB Execution # 17min
31 +- label: ":nvidia: (L4) EPLB Execution"
32 key: eplb-execution
33 timeout_in_minutes: 25
34 working_dir: "/vllm-workspace/tests"
@@ -39,7 +40,7 @@ steps:
40 - pytest -v -s distributed/test_eplb_execute.py
41 - pytest -v -s distributed/test_eplb_spec_decode.py
42
42 -- label: Elastic EP Scaling Test
43 +- label: ":nvidia: (H100) Elastic EP Scaling"
44 key: elastic-ep-scaling-test
45 timeout_in_minutes: 30
46 device: h100
.buildkite/test_areas/fault_tolerance.yaml
+1 -1
@@ -2,7 +2,7 @@ group: Fault Tolerance
2 depends_on:
3 - image-build
4 steps:
5 -- label: Fault Tolerance E2E (2xH100)
5 +- label: ":nvidia: (H100) Fault Tolerance E2E"
6 key: fault-tolerance-e2e-2xh100
7 timeout_in_minutes: 35
8 device: h100
.buildkite/test_areas/jit_monitor.yaml
+1 -1
@@ -2,7 +2,7 @@ group: JIT Monitor
2 depends_on:
3 - image-build
4 steps:
5 -- label: No Runtime JITs e2e tests
5 +- label: ":nvidia: (H200) No Runtime JITs E2E"
6 key: jit-monitor-no-runtime-jit
7 device: h200_35gb
8 timeout_in_minutes: 45
.buildkite/test_areas/kernels.yaml
+24 -21
@@ -2,7 +2,7 @@ group: Kernels
2 depends_on:
3 - image-build
4 steps:
5 -- label: vLLM IR Tests
5 +- label: ":nvidia: (H200) vLLM IR"
6 key: vllm-ir-tests
7 timeout_in_minutes: 35
8 device: h200_18gb
@@ -14,7 +14,7 @@ steps:
14 - pytest -v -s tests/ir
15 - pytest -v -s tests/kernels/ir
16
17 -- label: Kernels Core Operation Test
17 +- label: ":nvidia: (H200) Core Operation Kernels Shard %N"
18 device: h200_35gb
19 key: kernels-core-operation-test
20 timeout_in_minutes: 120
@@ -27,7 +27,7 @@ steps:
27 - pytest -v -s kernels/core --ignore=kernels/core/test_minimax_reduce_rms.py kernels/test_concat_mla_q.py kernels/test_fused_qk_norm_rope_gate.py --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
28 parallelism: 3
29
30 -- label: Kernels MiniMax Reduce RMS Test (2 GPUs)
30 +- label: ":nvidia: (H100) MiniMax Reduce RMS Kernels"
31 key: kernels-minimax-reduce-rms-test-2-gpus
32 timeout_in_minutes: 20
33 num_devices: 2
@@ -41,7 +41,7 @@ steps:
41 commands:
42 - pytest -v -s kernels/core/test_minimax_reduce_rms.py
43
44 -- label: Deepseek V4 Kernel Test (H100)
44 +- label: ":nvidia: (H100) DeepSeek V4 Kernels"
45 key: deepseek-v4-kernel-test-h100
46 timeout_in_minutes: 30
47 device: h100
@@ -54,7 +54,7 @@ steps:
54 - pytest -v -s kernels/test_fused_deepseek_v4_*.py
55 - pytest -v -s kernels/test_top_k_per_row.py
56
57 -- label: Deepseek V4 Kernel Test (B200)
57 +- label: ":nvidia: (B200) DeepSeek V4 Kernels"
58 key: deepseek-v4-kernel-test-b200
59 timeout_in_minutes: 20
60 device: b200-k8s
@@ -73,7 +73,7 @@ steps:
73 # Files with dedicated jobs elsewhere in this file are excluded via --ignore
74 # (test_kda, test_bf16x3_router_gemm_cutedsl and test_ll_bf16_gemm run in
75 # their own jobs / Kernels (B200)).
76 -- label: Kernels Root Misc Test (B200)
76 +- label: ":nvidia: (B200) Miscellaneous Kernels"
77 key: kernels-root-misc-test-b200
78 timeout_in_minutes: 45
79 device: b200-k8s
@@ -101,7 +101,7 @@ steps:
101 # BROKEN on main, pending kernel fixes (B200):
102 # test_shuffle_rows.py (1: test_shuffle_rows_edge_cases)
103
104 -- label: Kernels Attention Test %N
104 +- label: ":nvidia: (L4) Attention Kernels Shard %N"
105 key: kernels-attention-test
106 timeout_in_minutes: 65
107 source_file_dependencies:
@@ -116,6 +116,7 @@ steps:
116 parallelism: 2
117 mirror:
118 amd:
119 + label: ":amd: (MI300) Attention Kernels Shard %N"
120 dind: false
121 device: mi300_1
122 timeout_in_minutes: 90
@@ -130,7 +131,7 @@ steps:
131 - vllm/envs.py
132 - vllm/platforms/rocm.py
133
133 -- label: Kernels Attention DiffKV Test (H100)
134 +- label: ":nvidia: (H100) Attention DiffKV Kernels"
135 key: kernels-attention-diffkv-test-h100
136 timeout_in_minutes: 20
137 device: h100
@@ -143,7 +144,7 @@ steps:
144 commands:
145 - pytest -v -s kernels/attention/test_triton_unified_attention_diffkv.py
146
146 -- label: Kernels FlashMLA Test (H100)
147 +- label: ":nvidia: (H100) FlashMLA Kernels"
148 key: kernels-flashmla-test-h100
149 timeout_in_minutes: 25
150 device: h100
@@ -161,7 +162,7 @@ steps:
162 - pytest -v -s kernels/attention/test_flashmla_sparse.py
163 - pytest -v -s kernels/attention/test_mla_cross_layer_kernel_equivalence.py
164
164 -- label: Kernels Quantization Test %N
165 +- label: ":nvidia: (L4) Quantization Kernels Shard %N"
166 key: kernels-quantization-test
167 timeout_in_minutes: 60
168 source_file_dependencies:
@@ -173,6 +174,7 @@ steps:
174 parallelism: 2
175 mirror:
176 amd:
177 + label: ":amd: (MI300) Quantization Kernels Shard %N"
178 dind: false
179 device: mi300_1
180 timeout_in_minutes: 120
@@ -189,7 +191,7 @@ steps:
191 depends_on:
192 - image-build-amd
193
192 -- label: Kernels MoE Test %N
194 +- label: ":nvidia: (L4) MoE Kernels Shard %N"
195 key: kernels-moe-test
196 timeout_in_minutes: 50
197 source_file_dependencies:
@@ -206,6 +208,7 @@ steps:
208 parallelism: 5
209 mirror:
210 amd:
211 + label: ":amd: (MI300) MoE Kernels Shard %N"
212 dind: false
213 device: mi300_1
214 timeout_in_minutes: 55
@@ -222,7 +225,7 @@ steps:
225 depends_on:
226 - image-build-amd
227
225 -- label: Kernels Mamba Test
228 +- label: ":nvidia: (H200) Mamba Kernels"
229 device: h200_35gb
230 key: kernels-mamba-test
231 timeout_in_minutes: 60
@@ -233,7 +236,7 @@ steps:
236 commands:
237 - pytest -v -s kernels/mamba
238
236 -- label: Kernels DeepGEMM Test (H100)
239 +- label: ":nvidia: (H100) DeepGEMM Kernels"
240 key: kernels-deepgemm-test-h100
241 timeout_in_minutes: 35
242 device: h100
@@ -260,7 +263,7 @@ steps:
263 - pytest -v -s kernels/attention/test_deepgemm_attention.py
264 - pytest -v -s quantization/test_cutlass_w4a16.py
265
263 -- label: Kernels (B200)
266 +- label: ":nvidia: (B200) Kernels"
267 key: kernels-b200
268 timeout_in_minutes: 80
269 working_dir: "/vllm-workspace/"
@@ -340,7 +343,7 @@ steps:
343 # e2e
344 - pytest -v -s tests/models/quantization/test_nvfp4.py
345
343 -- label: B12X Linear Kernels (DGX Spark) Nightly
346 +- label: ":nvidia: (DGX) Spark B12X Linear Kernels Nightly"
347 key: b12x-linear-kernels-dgx-spark-nightly
348 timeout_in_minutes: 30
349 device: dgx-spark
@@ -362,7 +365,7 @@ steps:
365 model_executor/test_b12x_warmup.py
366 kernels/quantization/test_block_fp8.py -k b12x
367
365 -- label: Kernels Helion Test
368 +- label: ":nvidia: (H100) Helion Kernels Shard %N"
369 key: kernels-helion-test
370 timeout_in_minutes: 115
371 device: h100
@@ -375,7 +378,7 @@ steps:
378 parallelism: 2
379
380
378 -- label: Kernels FP8 MoE Test (1xH100)
381 +- label: ":nvidia: (H100) FP8 MoE Kernels"
382 key: kernels-fp8-moe-test-1xh100
383 timeout_in_minutes: 40
384 device: h100
@@ -392,7 +395,7 @@ steps:
395 - pytest -v -s kernels/moe/test_triton_moe_no_act_mul.py
396 - pytest -v -s kernels/moe/test_triton_moe_ptpc_fp8.py
397
395 -- label: Kernels FP8 MoE Test (2xH100)
398 +- label: ":nvidia: (H100) DeepEP FP8 MoE Kernels"
399 key: kernels-fp8-moe-test-2xh100
400 timeout_in_minutes: 45
401 device: h100
@@ -402,7 +405,7 @@ steps:
405 - pytest -v -s kernels/moe/test_deepep_deepgemm_moe.py
406 - pytest -v -s kernels/moe/test_deepep_moe.py
407
405 -- label: Kernels Fp4 MoE Test (B200)
408 +- label: ":nvidia: (B200) FP4 MoE Kernels"
409 key: kernels-fp4-moe-test-b200
410 timeout_in_minutes: 25
411 device: b200-k8s
@@ -415,7 +418,7 @@ steps:
418 - pytest -v -s kernels/moe/test_ocp_mx_moe.py
419
420
418 -- label: Kernels FusedMoE Layer Test (2 H100s)
421 +- label: ":nvidia: (H100) FusedMoE Layer Kernels"
422 key: kernels-fusedmoe-layer-test-2-h100s
423 timeout_in_minutes: 30
424 device: h100
@@ -432,7 +435,7 @@ steps:
435 - pytest -v -s kernels/moe/test_moe_layer.py
436
437
435 -- label: Kernels FusedMoE Layer Test (2 B200s)
438 +- label: ":nvidia: (B200) FusedMoE Layer Kernels"
439 key: kernels-fusedmoe-layer-test-2-b200s
440 timeout_in_minutes: 90
441 device: b200-k8s
.buildkite/test_areas/lm_eval.yaml
+28 -26
@@ -2,7 +2,7 @@ group: LM Eval
2 depends_on:
3 - image-build
4 steps:
5 -- label: LM Eval Small Models
5 +- label: ":nvidia: (H200) LM Eval Small Models"
6 device: h200_35gb
7 key: lm-eval-small-models
8 timeout_in_minutes: 45
@@ -14,6 +14,7 @@ steps:
14 - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-small.txt
15 mirror:
16 amd:
17 + label: ":amd: (MI300) LM Eval Small Models"
18 dind: false
19 device: mi300_1
20 timeout_in_minutes: 45
@@ -42,7 +43,7 @@ steps:
43 # - export VLLM_WORKER_MULTIPROC_METHOD=spawn
44 # - pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-large.txt --tp-size=4
45
45 -- label: LM Eval Large Models (4xH100)
46 +- label: ":nvidia: (H100) LM Eval Large Models"
47 key: lm-eval-large-models-4xh100
48 device: h100
49 optional: true
@@ -55,7 +56,7 @@ steps:
56 - export VLLM_USE_DEEP_GEMM=0 # We found Triton is faster than DeepGEMM for H100
57 - pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-large-hopper.txt --tp-size=4
58
58 -- label: LM Eval Small Models (1xB200)
59 +- label: ":nvidia: (B200) LM Eval Small Models"
60 key: lm-eval-small-models-1xb200
61 timeout_in_minutes: 50
62 device: b200-k8s
@@ -66,7 +67,7 @@ steps:
67 commands:
68 - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-blackwell.txt
69
69 -- label: LM Eval Small Models Distributed (2xB200)
70 +- label: ":nvidia: (B200) LM Eval Small Models Distributed"
71 key: lm-eval-small-models-distributed-2xb200
72 timeout_in_minutes: 120
73 device: b200-k8s
@@ -79,7 +80,7 @@ steps:
80 commands:
81 - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-small-tp.txt
82
82 -- label: LM Eval PCP (4xB200)
83 +- label: ":nvidia: (B200) LM Eval PCP"
84 key: lm-eval-pcp-4xb200
85 timeout_in_minutes: 360
86 device: b200-k8s
@@ -101,7 +102,7 @@ steps:
102 commands:
103 - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-pcp.txt
104
104 -- label: LM Eval Spec Decode (4xB200)
105 +- label: ":nvidia: (B200) LM Eval Spec Decode"
106 key: lm-eval-spec-decode-4xb200
107 timeout_in_minutes: 120
108 device: b200-k8s
@@ -121,7 +122,7 @@ steps:
122 commands:
123 - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-spec-decode.txt
124
124 -- label: LM Eval Large Models EP (2xB200)
125 +- label: ":nvidia: (B200) LM Eval Large Models EP"
126 key: lm-eval-large-models-ep-2xb200
127 timeout_in_minutes: 60
128 device: b200-k8s
@@ -133,7 +134,7 @@ steps:
134 commands:
135 - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-blackwell-ep.txt
136
136 -- label: LM Eval Qwen3.5 Models (2xB200)
137 +- label: ":nvidia: (B200) LM Eval Qwen3.5 Models"
138 key: lm-eval-qwen3-5-models-2xb200
139 timeout_in_minutes: 45
140 device: b200-k8s
@@ -150,7 +151,7 @@ steps:
151 commands:
152 - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-qwen35-blackwell.txt
153
153 -- label: LM Eval Large Models (8xH200)
154 +- label: ":nvidia: (H200) LM Eval Large Models"
155 key: lm-eval-large-models-8xh200
156 timeout_in_minutes: 50
157 device: h200
@@ -160,6 +161,7 @@ steps:
161 - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-h200.txt
162 mirror:
163 amd:
164 + label: ":amd: (MI300) LM Eval Large Models"
165 dind: false
166 device: mi300_8
167 timeout_in_minutes: 40
@@ -170,7 +172,7 @@ steps:
172 - export PYTORCH_ROCM_ARCH=gfx942 # Limit Quark compilation to save time
173 - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-mi3xx.txt
174
173 -- label: MoE Refactor Integration Test (H100 - TEMPORARY)
175 +- label: ":nvidia: (H100) MoE Refactor Integration TEMPORARY"
176 key: moe-refactor-integration-test-h100-temporary
177 device: h100
178 optional: true
@@ -178,7 +180,7 @@ steps:
180 commands:
181 - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/moe-refactor/config-h100.txt
182
181 -- label: MoE Refactor Integration Test (B200 - TEMPORARY) %N
183 +- label: ":nvidia: (B200) MoE Refactor Integration TEMPORARY Shard %N"
184 key: moe-refactor-integration-test-b200-temporary
185 device: b200-k8s
186 optional: true
@@ -187,7 +189,7 @@ steps:
189 commands:
190 - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/moe-refactor/config-b200-shard-$$BUILDKITE_PARALLEL_JOB.txt
191
190 -- label: MoE Refactor Integration Test (B200 DP - TEMPORARY)
192 +- label: ":nvidia: (B200) MoE Refactor DP Integration TEMPORARY"
193 key: moe-refactor-integration-test-b200-dp-temporary
194 device: b200-k8s
195 optional: true
@@ -195,7 +197,7 @@ steps:
197 commands:
198 - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/moe-refactor-dp-ep/config-b200.txt
199
198 -- label: LM Eval Humming f16 (A100 - TEMPORARY) %N
200 +- label: ":nvidia: (A100) LM Eval Humming FP16 TEMPORARY Shard %N"
201 key: lm-eval-humming-f16-a100
202 timeout_in_minutes: 75
203 device: a100
@@ -211,7 +213,7 @@ steps:
213 commands:
214 - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/humming/config-a100-shard-$$BUILDKITE_PARALLEL_JOB.txt
215
214 -- label: LM Eval Humming Act int8 (A100 - TEMPORARY)
216 +- label: ":nvidia: (A100) LM Eval Humming Activation INT8 TEMPORARY"
217 key: lm-eval-humming-act-a100
218 timeout_in_minutes: 45
219 device: a100
@@ -226,7 +228,7 @@ steps:
228 commands:
229 - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/humming/config-act-int8.txt
230
229 -- label: LM Eval Humming f16 (H100 - TEMPORARY) %N
231 +- label: ":nvidia: (H100) LM Eval Humming FP16 TEMPORARY Shard %N"
232 key: lm-eval-humming-f16-h100
233 timeout_in_minutes: 70
234 device: h100
@@ -242,7 +244,7 @@ steps:
244 commands:
245 - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/humming/config-h100-shard-$$BUILDKITE_PARALLEL_JOB.txt
246
245 -- label: LM Eval Humming Act fp8/int8 (H100 - TEMPORARY)
247 +- label: ":nvidia: (H100) LM Eval Humming Activation FP8/INT8 TEMPORARY"
248 key: lm-eval-humming-act-h100
249 timeout_in_minutes: 70
250 device: h100
@@ -258,7 +260,7 @@ steps:
260 - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/humming/config-act-fp8.txt
261 # - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/humming/config-act-int8.txt
262
261 -- label: LM Eval Humming f16 (B200 - TEMPORARY)
263 +- label: ":nvidia: (B200) LM Eval Humming FP16 TEMPORARY"
264 key: lm-eval-humming-f16-b200
265 timeout_in_minutes: 50
266 device: b200-k8s
@@ -273,7 +275,7 @@ steps:
275 commands:
276 - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/humming/config.txt
277
276 -- label: LM Eval Humming Act fp8/int8 (B200 - TEMPORARY)
278 +- label: ":nvidia: (B200) LM Eval Humming Activation FP8/INT8 TEMPORARY"
279 key: lm-eval-humming-act-b200
280 timeout_in_minutes: 50
281 device: b200-k8s
@@ -289,7 +291,7 @@ steps:
291 - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/humming/config-act-fp8.txt
292 - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/humming/config-act-int8.txt
293
292 -- label: LM Eval TurboQuant KV Cache
294 +- label: ":nvidia: (H200) LM Eval TurboQuant KV Cache"
295 key: lm-eval-turboquant-kv-cache
296 timeout_in_minutes: 55
297 device: h200_18gb
@@ -301,7 +303,7 @@ steps:
303 commands:
304 - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/models-turboquant.txt
305
304 -- label: GPQA Eval (GPT-OSS) (2xH100)
306 +- label: ":nvidia: (H100) GPQA Eval (GPT-OSS)"
307 key: gpqa-eval-gpt-oss-2xh100
308 timeout_in_minutes: 35
309 device: h100
@@ -315,7 +317,7 @@ steps:
317 - uv pip install --system 'gpt-oss[eval]==0.0.5'
318 - pytest -s -v evals/gpt_oss/test_gpqa_correctness.py --config-list-file=configs/models-h100.txt
319
318 -- label: GPQA Eval (GPT-OSS) (2xB200)
320 +- label: ":nvidia: (B200) GPQA Eval (GPT-OSS)"
321 key: gpqa-eval-gpt-oss-2xb200
322 timeout_in_minutes: 30
323 device: b200-k8s
@@ -329,7 +331,7 @@ steps:
331 - uv pip install --system 'gpt-oss[eval]==0.0.5'
332 - pytest -s -v evals/gpt_oss/test_gpqa_correctness.py --config-list-file=configs/models-b200.txt
333
332 -- label: GPQA Eval (GPT-OSS) (DGX Spark)
334 +- label: ":nvidia: (DGX) Spark GPQA Eval (GPT-OSS)"
335 key: gpqa-eval-gpt-oss-spark
336 timeout_in_minutes: 35
337 device: dgx-spark
@@ -345,7 +347,7 @@ steps:
347 - uv pip install --system 'gpt-oss[eval]==0.0.5'
348 - pytest -s -v evals/gpt_oss/test_gpqa_correctness.py --config-list-file=configs/models-spark.txt
349
348 -- label: LM Eval KV-Offload (1xH200)
350 +- label: ":nvidia: (H200) LM Eval KV-Offload"
351 key: kv-offload-small
352 timeout_in_minutes: 30
353 device: h200_35gb
@@ -358,7 +360,7 @@ steps:
360 commands:
361 - pytest -s -v evals/gsm8k/test_gsm8k_offloading.py -k "nemotron-h-8b or gemma-4-e4b-it"
362
361 -- label: LM Eval KV-Offload (2xH100)
363 +- label: ":nvidia: (H100) LM Eval KV-Offload Medium"
364 key: kv-offload-medium
365 timeout_in_minutes: 45
366 device: h100
@@ -372,7 +374,7 @@ steps:
374 commands:
375 - pytest -s -v evals/gsm8k/test_gsm8k_offloading.py -k "qwen3.5-35b or deepseek-v2-lite"
376
375 -- label: LM Eval KV-Offload (4xH100)
377 +- label: ":nvidia: (H100) LM Eval KV-Offload Large"
378 key: kv-offload-large
379 timeout_in_minutes: 40
380 device: h100
@@ -386,7 +388,7 @@ steps:
388 commands:
389 - pytest -s -v evals/gsm8k/test_gsm8k_offloading.py -k "deepseek-v4-flash"
390
389 -- label: MRCR Eval Small Models
391 +- label: ":nvidia: (H200) MRCR Eval Small Models"
392 device: h200_35gb
393 timeout_in_minutes: 25
394 source_file_dependencies:
.buildkite/test_areas/lora.yaml
+3 -2
@@ -2,7 +2,7 @@ group: LoRA
2 depends_on:
3 - image-build
4 steps:
5 -- label: LoRA %N
5 +- label: ":nvidia: (H200) LoRA Shard %N"
6 device: h200_35gb
7 key: lora
8 timeout_in_minutes: 40
@@ -14,6 +14,7 @@ steps:
14 parallelism: 4
15 mirror:
16 amd:
17 + label: ":amd: (MI300) LoRA Shard %N"
18 dind: false
19 device: mi300_1
20 working_dir: "/vllm-workspace/tests"
@@ -26,7 +27,7 @@ steps:
27 - image-build-amd
28
29
29 -- label: LoRA TP (Distributed)
30 +- label: ":nvidia: (L4) LoRA TP (Distributed)"
31 key: lora-tp-distributed
32 timeout_in_minutes: 60
33 num_devices: 4
.buildkite/test_areas/misc.yaml
+20 -15
@@ -2,7 +2,7 @@ group: Miscellaneous
2 depends_on:
3 - image-build
4 steps:
5 -- label: V1 Sample + Logits
5 +- label: ":nvidia: (H200) V1 Sample + Logits"
6 key: v1-sample-logits
7 timeout_in_minutes: 83
8 device: h200_18gb
@@ -33,13 +33,14 @@ steps:
33 - pytest -v -s v1/test_outputs.py
34 mirror:
35 amd:
36 + label: ":amd: (MI300) V1 Sample + Logits"
37 dind: false
38 device: mi300_1
39 timeout_in_minutes: 70
40 depends_on:
41 - image-build-amd
42
42 -- label: V1 Core + KV + Metrics
43 +- label: ":nvidia: (H200) V1 Core + KV + Metrics"
44 device: h200_35gb
45 key: v1-core-kv-metrics
46 timeout_in_minutes: 80
@@ -90,13 +91,14 @@ steps:
91 - pytest -v -s entrypoints/openai/correctness/test_lmeval.py::test_lm_eval_accuracy_v1_engine
92 mirror:
93 amd:
94 + label: ":amd: (MI300) V1 Core + KV + Metrics"
95 dind: false
96 device: mi300_1
97 timeout_in_minutes: 65
98 depends_on:
99 - image-build-amd
100
99 -- label: V1 Others (CPU)
101 +- label: ":computer: (CPU) V1 Others"
102 key: v1-others-cpu
103 depends_on:
104 - image-build-cpu
@@ -130,7 +132,7 @@ steps:
132 - pytest -v -s -m 'cpu_test' v1/ec_connector/unit
133 - pytest -v -s -m 'cpu_test' v1/metrics
134
133 -- label: Extract Hidden States Integration
135 +- label: ":nvidia: (H200) Extract Hidden States Integration"
136 key: extract-hidden-states-integration
137 timeout_in_minutes: 20
138 device: h200_18gb
@@ -144,7 +146,7 @@ steps:
146 - export VLLM_WORKER_MULTIPROC_METHOD=spawn
147 - pytest -v -s v1/kv_connector/extract_hidden_states_integration
148
147 -- label: Extract Hidden States Integration (2 GPUs)
149 +- label: ":nvidia: (L4) Extract Hidden States Integration"
150 key: extract-hidden-states-integration-2-gpus
151 timeout_in_minutes: 20
152 num_devices: 2
@@ -158,7 +160,7 @@ steps:
160 - export VLLM_WORKER_MULTIPROC_METHOD=spawn
161 - pytest -v -s -m 'distributed' v1/kv_connector/extract_hidden_states_integration
162
161 -- label: Regression
163 +- label: ":nvidia: (H200) Regression"
164 key: regression
165 timeout_in_minutes: 30
166 device: h200_18gb
@@ -181,7 +183,7 @@ steps:
183 - pytest -v -s test_regression.py
184 working_dir: "/vllm-workspace/tests" # optional
185
184 -- label: Examples
186 +- label: ":nvidia: (H200) Examples"
187 device: h200_35gb
188 key: examples
189 timeout_in_minutes: 40
@@ -215,6 +217,7 @@ steps:
217 - python3 features/speculative_decoding/spec_decode_offline.py --test --method eagle3 --num_spec_tokens 3 --dataset-name hf --dataset-path philschmid/mt-bench --num-prompts 80 --temp 0 --top-p 1.0 --top-k -1 --tp 1 --enable-chunked-prefill --max-model-len 1536
218 mirror:
219 amd:
220 + label: ":amd: (MI300) Examples"
221 dind: false
222 device: mi300_1
223 timeout_in_minutes: 75
@@ -226,7 +229,7 @@ steps:
229 depends_on:
230 - image-build-amd
231
229 -- label: Metrics, Tracing (2 GPUs)
232 +- label: ":nvidia: (L4) Metrics, Tracing"
233 key: metrics-tracing-2-gpus
234 timeout_in_minutes: 25
235 num_devices: 2
@@ -256,13 +259,14 @@ steps:
259 - pytest -v -s tracing
260 mirror:
261 amd:
262 + label: ":amd: (MI300) Metrics, Tracing"
263 dind: false
264 device: mi300_2
265 timeout_in_minutes: 30
266 depends_on:
267 - image-build-amd
268
265 -- label: Python-only Installation
269 +- label: ":nvidia: (H200) Python-only Installation"
270 key: python-only-installation
271 depends_on: ~
272 optional: true
@@ -275,6 +279,7 @@ steps:
279 - bash standalone_tests/python_only_compile.sh
280 mirror:
281 amd:
282 + label: ":amd: (MI300) Python-only Installation"
283 dind: false
284 device: mi300_1
285 timeout_in_minutes: 55
@@ -286,7 +291,7 @@ steps:
291 - setup.py
292 - vllm/platforms/rocm.py
293
289 -- label: Async Engine, Inputs, Utils, Worker
294 +- label: ":nvidia: (H200) Async Engine, Inputs, Utils, Worker"
295 device: h200_35gb
296 key: async-engine-inputs-utils-worker
297 timeout_in_minutes: 25
@@ -313,7 +318,7 @@ steps:
318 - pytest -v -s -m 'not cpu_test' multimodal
319 - pytest -v -s utils_
320
316 -- label: Async Engine, Inputs, Utils, Worker, Config (CPU)
321 +- label: ":computer: (CPU) Async Engine, Inputs, Utils, Worker, Config"
322 key: async-engine-inputs-utils-worker-config-cpu
323 depends_on:
324 - image-build-cpu
@@ -377,7 +382,7 @@ steps:
382 - pytest -v -s transformers_utils
383 - pytest -v -s config
384
380 -- label: Batch Invariance (A100)
385 +- label: ":nvidia: (A100) Batch Invariance"
386 key: batch-invariance-a100
387 timeout_in_minutes: 60
388 device: a100
@@ -391,7 +396,7 @@ steps:
396 - pytest -v -s v1/determinism/test_batch_invariance.py
397 - VLLM_TEST_MODEL=deepseek-ai/DeepSeek-V2-Lite-Chat pytest -v -s v1/determinism/test_batch_invariance.py::test_v1_generation_is_deterministic_across_batch_sizes_with_needle -k TRITON_MLA
398
394 -- label: Batch Invariance (H100)
399 +- label: ":nvidia: (H100) Batch Invariance"
400 key: batch-invariance-h100
401 timeout_in_minutes: 60
402 device: h100
@@ -407,7 +412,7 @@ steps:
412 - VLLM_TEST_MODEL=deepseek-ai/DeepSeek-V2-Lite-Chat pytest -v -s v1/determinism/test_batch_invariance.py::test_v1_generation_is_deterministic_across_batch_sizes_with_needle -k TRITON_MLA
413 - VLLM_TEST_MODEL=Qwen/Qwen3-30B-A3B-Thinking-2507-FP8 pytest -v -s v1/determinism/test_batch_invariance.py::test_v1_generation_is_deterministic_across_batch_sizes_with_needle -k FLASH_ATTN
414
410 -- label: Batch Invariance (B200)
415 +- label: ":nvidia: (B200) Batch Invariance"
416 key: batch-invariance-b200
417 timeout_in_minutes: 45
418 device: b200-k8s
@@ -427,7 +432,7 @@ steps:
432 - pytest -v -s v1/determinism/test_cutlass_batch_invariance.py
433 - pytest -v -s v1/determinism/test_online_batch_invariance.py
434
430 -- label: Acceptance Length Test (Large Models) # optional
435 +- label: ":nvidia: (H200) Acceptance Length (Large Models)"
436 device: h200_35gb
437 key: acceptance-length-test-large-models
438 timeout_in_minutes: 20
.buildkite/test_areas/model_executor.yaml
+2 -1
@@ -2,7 +2,7 @@ group: Model Executor
2 depends_on:
3 - image-build
4 steps:
5 -- label: Model Executor
5 +- label: ":nvidia: (H200) Model Executor"
6 device: h200_35gb
7 key: model-executor
8 timeout_in_minutes: 60
@@ -28,6 +28,7 @@ steps:
28 - pytest -v -s entrypoints/openai/completion/test_tensorizer_entrypoint.py --timeout=900 --timeout-method=thread
29 mirror:
30 amd:
31 + label: ":amd: (MI300) Model Executor"
32 dind: false
33 device: mi300_1
34 timeout_in_minutes: 60
.buildkite/test_areas/model_runner_v2.yaml
+5 -5
@@ -2,7 +2,7 @@ group: Model Runner V2
2 depends_on:
3 - image-build
4 steps:
5 -- label: Model Runner V2 Core Tests
5 +- label: ":nvidia: (H200) Model Runner V2 Core"
6 device: h200_35gb
7 key: model-runner-v2-core-tests
8 timeout_in_minutes: 35
@@ -24,7 +24,7 @@ steps:
24 # Temporary hack filter to exclude ngram spec decoding based tests.
25 - 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"
26
27 -- label: Model Runner V2 Examples
27 +- label: ":nvidia: (H200) Model Runner V2 Examples"
28 device: h200_35gb
29 key: model-runner-v2-examples
30 timeout_in_minutes: 35
@@ -61,7 +61,7 @@ steps:
61 # https://github.com/vllm-project/vllm/pull/26682 uses slightly more memory in PyTorch 2.9+ causing this test to OOM in 1xL4 GPU
62 - python3 features/speculative_decoding/spec_decode_offline.py --test --method eagle3 --num_spec_tokens 3 --dataset-name hf --dataset-path philschmid/mt-bench --num-prompts 80 --temp 0 --top-p 1.0 --top-k -1 --tp 1 --enable-chunked-prefill --max-model-len 1536
63
64 -- label: Model Runner V2 Distributed (2 GPUs)
64 +- label: ":nvidia: (L4) Model Runner V2 Distributed"
65 key: model-runner-v2-distributed-2-gpus
66 timeout_in_minutes: 30
67 working_dir: "/vllm-workspace/tests"
@@ -82,7 +82,7 @@ steps:
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
85 -- label: Model Runner V2 Pipeline Parallelism (4 GPUs)
85 +- label: ":nvidia: (L4) Model Runner V2 Pipeline Parallelism"
86 key: model-runner-v2-pipeline-parallelism-4-gpus
87 timeout_in_minutes: 50
88 working_dir: "/vllm-workspace/tests"
@@ -100,7 +100,7 @@ steps:
100 - pytest -v -s distributed/test_pp_cudagraph.py -k "not ray"
101 - pytest -v -s v1/distributed/test_pp_dp_v2.py
102
103 -- label: Model Runner V2 Spec Decode
103 +- label: ":nvidia: (H200) Model Runner V2 Spec Decode"
104 device: h200_35gb
105 key: model-runner-v2-spec-decode
106 timeout_in_minutes: 50
.buildkite/test_areas/models_basic.yaml
+7 -6
@@ -2,7 +2,7 @@ group: Models - Basic
2 depends_on:
3 - image-build
4 steps:
5 -- label: Basic Models Tests (Initialization)
5 +- label: ":nvidia: (H200) Basic Models (Initialization)"
6 key: basic-models-tests-initialization
7 timeout_in_minutes: 25
8 device: h200_18gb
@@ -15,7 +15,7 @@ steps:
15 # Run a subset of model initialization tests
16 - pytest -v -s models/test_initialization.py::test_can_initialize_small_subset
17
18 -- label: Basic Models Tests (Extra Initialization) %N
18 +- label: ":nvidia: (H200) Basic Models (Extra Initialization) Shard %N"
19 device: h200_35gb
20 key: basic-models-tests-extra-initialization
21 timeout_in_minutes: 100
@@ -30,7 +30,7 @@ steps:
30 - pytest -v -s models/test_initialization.py -k 'not test_can_initialize_small_subset' --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
31 parallelism: 4
32
33 -- label: Basic Models Tests (Other)
33 +- label: ":nvidia: (H200) Basic Models (Other)"
34 device: h200_35gb
35 key: basic-models-tests-other
36 timeout_in_minutes: 35
@@ -44,13 +44,14 @@ steps:
44 - pytest -v -s models/test_terratorch.py models/transformers/test_backend.py models/test_registry.py
45 mirror:
46 amd:
47 + label: ":amd: (MI300) Basic Models (Other)"
48 dind: false
49 device: mi300_1
50 timeout_in_minutes: 50
51 depends_on:
52 - image-build-amd
53
53 -- label: Inkling Unit Tests (B200)
54 +- label: ":nvidia: (B200) Inkling"
55 key: inkling-unit-tests-b200
56 timeout_in_minutes: 40
57 device: b200-k8s
@@ -63,7 +64,7 @@ steps:
64 # FA4 kernel tests require SM100; the suite skips them elsewhere.
65 - pytest -v -s models/inkling
66
66 -- label: Kimi K3 Unit Tests (B200)
67 +- label: ":nvidia: (B200) Kimi K3"
68 key: kimi-k3-unit-tests-b200
69 timeout_in_minutes: 40
70 device: b200-k8s
@@ -77,7 +78,7 @@ steps:
78 # The native NVIDIA Kimi K3 kernels require the SM100 family.
79 - pytest -v -s models/kimi_k3 kernels/attention/test_kimi_k3_mla_fused_epilogue.py kernels/test_bf16_skinny_gemm.py
80
80 -- label: Basic Models Test (Other CPU) # 5min
81 +- label: ":computer: (CPU) Basic Models Other"
82 key: basic-models-test-other-cpu
83 depends_on:
84 - image-build-cpu
.buildkite/test_areas/models_distributed.yaml
+1 -1
@@ -2,7 +2,7 @@ group: Models - Distributed
2 depends_on:
3 - image-build
4 steps:
5 -- label: Distributed Model Tests (2 GPUs)
5 +- label: ":nvidia: (L4) Distributed Models"
6 key: distributed-model-tests-2-gpus
7 timeout_in_minutes: 60
8 working_dir: "/vllm-workspace/tests"
.buildkite/test_areas/models_language.yaml
+12 -8
@@ -2,7 +2,7 @@ group: Models - Language
2 depends_on:
3 - image-build
4 steps:
5 -- label: Language Models Tests (Standard)
5 +- label: ":nvidia: (H200) Language Models (Standard)"
6 key: language-models-tests-standard
7 timeout_in_minutes: 30
8 device: h200_18gb
@@ -16,13 +16,14 @@ steps:
16 - pytest -v -s models/language -m 'core_model and (not slow_test)'
17 mirror:
18 amd:
19 + label: ":amd: (MI300) Language Models (Standard)"
20 dind: false
21 device: mi300_1
22 timeout_in_minutes: 45
23 depends_on:
24 - image-build-amd
25
25 -- label: Language Models Tests (Extra Standard) %N
26 +- label: ":nvidia: (H200) Language Models (Extra Standard) Shard %N"
27 device: h200_35gb
28 key: language-models-tests-extra-standard
29 timeout_in_minutes: 40
@@ -39,6 +40,7 @@ steps:
40 parallelism: 2
41 mirror:
42 amd:
43 + label: ":amd: (MI300) Language Models (Extra Standard) Shard %N"
44 dind: false
45 device: mi300_1
46 timeout_in_minutes: 40
@@ -55,7 +57,7 @@ steps:
57 - tests/models/language/pooling/test_classification.py
58 - vllm/_aiter_ops.py
59 - vllm/platforms/rocm.py
58 -- label: Language Models Tests (Hybrid) %N
60 +- label: ":nvidia: (H200) Language Models (Hybrid) Shard %N"
61 device: h200_35gb
62 key: language-models-tests-hybrid
63 timeout_in_minutes: 65
@@ -72,6 +74,7 @@ steps:
74 parallelism: 2
75 mirror:
76 amd:
77 + label: ":amd: (MI300) Language Models (Hybrid) Shard %N"
78 dind: false
79 device: mi300_1
80 timeout_in_minutes: 60
@@ -85,7 +88,7 @@ steps:
88 # Granite 4 hybrid generation is sensitive to hardware-specific Triton SSD
89 # autotuning (https://github.com/vllm-project/vllm/issues/25194). Keep this one
90 # correctness test on L4 until its H200 output matches the Transformers reference.
88 -- label: Language Models Tests (Granite L4 Compatibility)
91 +- label: ":nvidia: (L4) Granite Language Model Compatibility"
92 key: language-models-tests-granite-l4-compatibility
93 timeout_in_minutes: 65
94 source_file_dependencies:
@@ -97,7 +100,7 @@ steps:
100 - CAUSAL_CONV1D_FORCE_BUILD=TRUE uv pip install --system --no-build-isolation 'git+https://github.com/Dao-AILab/causal-conv1d@v1.6.0'
101 - pytest -v -s models/language/generation -m hybrid_model -k 'granite-4.0-tiny-preview'
102
100 -- label: Language Models Test (Extended Generation) # 80min
103 +- label: ":nvidia: (H200) Language Models (Extended Generation)"
104 device: h200_35gb
105 key: language-models-test-extended-generation
106 timeout_in_minutes: 80
@@ -112,7 +115,7 @@ steps:
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
115 -- label: Language Models Test (PPL)
118 +- label: ":nvidia: (H200) Language Models (PPL)"
119 key: language-models-test-ppl
120 timeout_in_minutes: 30
121 device: h200_18gb
@@ -124,7 +127,7 @@ steps:
127 commands:
128 - pytest -v -s models/language/generation_ppl_test
129
127 -- label: Language Models Test (Extended Pooling) %N
130 +- label: ":nvidia: (H200) Language Models (Extended Pooling) Shard %N"
131 device: h200_35gb
132 key: language-models-test-extended-pooling
133 timeout_in_minutes: 120
@@ -138,6 +141,7 @@ steps:
141 - pytest -v -s models/language/pooling -m 'not core_model' --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
142 mirror:
143 amd:
144 + label: ":amd: (MI300) Language Models (Extended Pooling) Shard %N"
145 dind: false
146 device: mi300_1
147 timeout_in_minutes: 95
@@ -147,7 +151,7 @@ steps:
151 commands:
152 - pytest -v -s models/language/pooling -m 'not core_model' --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
153
150 -- label: Language Models Test (MTEB)
154 +- label: ":nvidia: (H200) Language Models (MTEB)"
155 key: language-models-test-mteb
156 timeout_in_minutes: 68
157 device: h200_18gb
.buildkite/test_areas/models_multimodal.yaml
+20 -12
@@ -2,7 +2,7 @@ group: Models - Multimodal
2 depends_on:
3 - image-build
4 steps:
5 -- label: "Multi-Modal Models (Standard) 1: qwen2"
5 +- label: ":nvidia: (H200) Multimodal Models (Standard) 1: qwen2"
6 key: multi-modal-models-standard-1-qwen2
7 timeout_in_minutes: 68
8 device: h200_18gb
@@ -15,13 +15,14 @@ steps:
15 - pytest -v -s models/multimodal/generation/test_ultravox.py -m core_model
16 mirror:
17 amd:
18 + label: ":amd: (MI300) Multimodal Models (Standard) 1: qwen2"
19 dind: false
20 device: mi300_1
21 timeout_in_minutes: 65
22 depends_on:
23 - image-build-amd
24
24 -- label: "Multi-Modal Models (Standard) 2: qwen3 + gemma"
25 +- label: ":nvidia: (H200) Multimodal Models (Standard) 2: qwen3 + gemma"
26 key: multi-modal-models-standard-2-qwen3-gemma
27 timeout_in_minutes: 75
28 device: h200_18gb
@@ -35,13 +36,14 @@ steps:
36 - pytest -v -s models/multimodal/generation/test_qwen2_5_vl.py -m core_model
37 mirror:
38 amd:
39 + label: ":amd: (MI300) Multimodal Models (Standard) 2: qwen3 + gemma"
40 dind: false
41 device: mi300_1
42 timeout_in_minutes: 55
43 depends_on:
44 - image-build-amd
45
44 -- label: "Multi-Modal Models (Standard) 3: llava + qwen2_vl"
46 +- label: ":nvidia: (H200) Multimodal Models (Standard) 3: llava + qwen2_vl"
47 device: h200_35gb
48 key: multi-modal-models-standard-3-llava-qwen2-vl
49 timeout_in_minutes: 40
@@ -54,12 +56,13 @@ steps:
56 - pytest -v -s models/multimodal/generation/test_qwen2_vl.py -m core_model
57 mirror:
58 amd:
59 + label: ":amd: (MI250) Multimodal Models (Standard) 3: llava + qwen2_vl"
60 device: mi250_1
61 timeout_in_minutes: 55
62 depends_on:
63 - image-build-amd
64
62 -- label: "Multi-Modal Models (Standard) 4: other + whisper"
65 +- label: ":nvidia: (H200) Multimodal Models (Standard) 4: other + whisper"
66 device: h200_35gb
67 key: multi-modal-models-standard-4-other-whisper
68 timeout_in_minutes: 75
@@ -74,13 +77,14 @@ steps:
77 - cd .. && VLLM_WORKER_MULTIPROC_METHOD=spawn pytest -v -s tests/models/multimodal/generation/test_whisper.py -m core_model # Otherwise, mp_method="spawn" doesn't work
78 mirror:
79 amd:
80 + label: ":amd: (MI300) Multimodal Models (Standard) 4: other + whisper"
81 dind: false
82 device: mi300_1
83 timeout_in_minutes: 50
84 depends_on:
85 - image-build-amd
86
83 -- label: Multi-Modal Processor (CPU) %N
87 +- label: ":computer: (CPU) Multimodal Processor Shard %N"
88 key: multi-modal-processor-cpu
89 depends_on:
90 - image-build-cpu
@@ -95,7 +99,7 @@ steps:
99 - pytest -v -s models/multimodal/processing --ignore models/multimodal/processing/test_tensor_schema.py --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
100 parallelism: 4
101
98 -- label: Multi-Modal Processor # 44min
102 +- label: ":nvidia: (H200) Multimodal Processor"
103 key: multi-modal-processor
104 timeout_in_minutes: 98
105 device: h200_18gb
@@ -107,7 +111,7 @@ steps:
111 commands:
112 - pytest -v -s models/multimodal/processing/test_tensor_schema.py
113
110 -- label: Multi-Modal Accuracy Eval (Small Models) # 50min
114 +- label: ":nvidia: (H200) Multimodal Accuracy Eval (Small Models)"
115 device: h200_35gb
116 key: multi-modal-accuracy-eval-small-models
117 timeout_in_minutes: 30
@@ -120,6 +124,7 @@ steps:
124 - pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-mm-small.txt --tp-size=1
125 mirror:
126 amd:
127 + label: ":amd: (MI300) Multimodal Accuracy Eval (Small Models)"
128 dind: false
129 device: mi300_1
130 timeout_in_minutes: 35
@@ -132,7 +137,7 @@ steps:
137 - vllm/platforms/rocm.py
138 - vllm/model_executor/model_loader/
139
135 -- label: Multi-Modal Models (Extended Generation 1)
140 +- label: ":nvidia: (H200) Multimodal Models (Extended Generation 1)"
141 device: h200_35gb
142 key: multi-modal-models-extended-generation-1
143 optional: true
@@ -146,13 +151,14 @@ steps:
151 - pytest -v -s models/multimodal/test_mapping.py
152 mirror:
153 amd:
154 + label: ":amd: (MI300) Multimodal Models (Extended Generation 1)"
155 dind: false
156 device: mi300_1
157 timeout_in_minutes: 90
158 depends_on:
159 - image-build-amd
160
155 -- label: Multi-Modal Models (PPL)
161 +- label: ":nvidia: (H200) Multimodal Models (PPL)"
162 device: h200_35gb
163 key: multi-modal-models-extended-ppl
164 optional: true
@@ -163,13 +169,14 @@ steps:
169 - pytest -v -s models/multimodal/generation_ppl_test/
170 mirror:
171 amd:
172 + label: ":amd: (MI300) Multimodal Models (PPL)"
173 dind: false
174 device: mi300_1
175 timeout_in_minutes: 90
176 depends_on:
177 - image-build-amd
178
172 -- label: Multi-Modal Models (Extended Generation 2) %N
179 +- label: ":nvidia: (H200) Multimodal Models (Extended Generation 2) Shard %N"
180 device: h200_35gb
181 key: multi-modal-models-extended-generation-2
182 parallelism: 4
@@ -181,7 +188,7 @@ steps:
188 commands:
189 - 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
190
184 -- label: Multi-Modal Models (Extended Generation 3)
191 +- label: ":nvidia: (H200) Multimodal Models (Extended Generation 3)"
192 device: h200_35gb
193 key: multi-modal-models-extended-generation-3
194 optional: true
@@ -192,7 +199,7 @@ steps:
199 commands:
200 - pytest -v -s models/multimodal/generation/test_common.py -m 'split(group=1) and not core_model'
201
195 -- label: Multi-Modal Models (Extended Pooling)
202 +- label: ":nvidia: (H200) Multimodal Models (Extended Pooling)"
203 key: multi-modal-models-extended-pooling
204 optional: true
205 device: h200_18gb
@@ -204,6 +211,7 @@ steps:
211 - pytest -v -s models/multimodal/pooling -m 'not core_model'
212 mirror:
213 amd:
214 + label: ":amd: (MI300) Multimodal Models (Extended Pooling)"
215 dind: false
216 device: mi300_1
217 timeout_in_minutes: 60
.buildkite/test_areas/plugins.yaml
+4 -4
@@ -2,7 +2,7 @@ group: Plugins
2 depends_on:
3 - image-build
4 steps:
5 -- label: Plugin Tests (2 GPUs)
5 +- label: ":nvidia: (L4) Plugin Integration"
6 key: plugin-tests-2-gpus
7 timeout_in_minutes: 35
8 working_dir: "/vllm-workspace/tests"
@@ -51,7 +51,7 @@ steps:
51 - pytest -v -s plugins_tests/lora_resolvers # unit tests for in-tree lora resolver plugins
52
53
54 -- label: GGUF Plugin
54 +- label: ":nvidia: (H200) GGUF Plugin"
55 key: gguf-plugin
56 device: h200_18gb
57 timeout_in_minutes: 30
@@ -64,7 +64,7 @@ steps:
64 - pip install "vllm-gguf-plugin >= 0.0.2"
65 - pytest -v -s plugins_tests/gguf
66
67 -- label: BitsAndBytes Plugin
67 +- label: ":nvidia: (H200) BitsAndBytes Plugin"
68 key: bitsandbytes-plugin
69 device: h200_18gb
70 timeout_in_minutes: 30
@@ -81,7 +81,7 @@ steps:
81 - pip install "vllm-bnb-plugin >= 0.0.1"
82 - pytest -v -s plugins_tests/bitsandbytes -m 'not distributed'
83
84 -- label: BitsAndBytes Plugin (2 GPUs)
84 +- label: ":nvidia: (L4) BitsAndBytes Plugin"
85 key: bitsandbytes-plugin-2-gpus
86 timeout_in_minutes: 15
87 working_dir: "/vllm-workspace/tests"
.buildkite/test_areas/pytorch.yaml
+7 -6
@@ -2,7 +2,7 @@ group: PyTorch
2 depends_on:
3 - image-build
4 steps:
5 -- label: PyTorch Compilation Unit Tests
5 +- label: ":nvidia: (H200) PyTorch Compilation"
6 device: h200_35gb
7 key: pytorch-compilation-unit-tests
8 timeout_in_minutes: 60
@@ -43,7 +43,7 @@ steps:
43 # (using -0 for proper path handling)
44 - "find compile/ -maxdepth 1 -name 'test_*.py' -print0 | xargs -0 -n1 -I{} pytest -s -v '{}'"
45
46 -- label: PyTorch Compilation Unit Tests (H100)
46 +- label: ":nvidia: (H200) PyTorch Compilation H100 Cases"
47 key: pytorch-compilation-unit-tests-h100
48 timeout_in_minutes: 30
49 device: h200_18gb
@@ -77,7 +77,7 @@ steps:
77 commands:
78 - "find compile/h100/ -name 'test_*.py' -print0 | xargs -0 -n1 -I{} pytest -s -v '{}'"
79
80 -- label: PyTorch Compilation Passes Unit Tests
80 +- label: ":nvidia: (L4) PyTorch Compilation Passes"
81 key: pytorch-compilation-passes-unit-tests
82 timeout_in_minutes: 45
83 source_file_dependencies:
@@ -110,7 +110,7 @@ steps:
110 commands:
111 - pytest -s -v compile/passes --ignore compile/passes/distributed
112
113 -- label: PyTorch Fullgraph Test
113 +- label: ":nvidia: (H200) PyTorch Fullgraph"
114 device: h200_35gb
115 key: pytorch-fullgraph-test
116 timeout_in_minutes: 90
@@ -151,7 +151,7 @@ steps:
151 # Hopper-only DeepSeek-V2-Lite cases in this file require two 29.3-GiB model
152 # instances and cannot fit a 35GB MIG slice. L4 retains the original coverage:
153 # those SM90 cases skip while the architecture-compatible cases still run.
154 -- label: PyTorch Fullgraph CUDAGraph (L4 Compatibility)
154 +- label: ":nvidia: (L4) PyTorch Fullgraph CUDAGraph Compatibility"
155 key: pytorch-fullgraph-cudagraph-l4-compatibility
156 timeout_in_minutes: 60
157 source_file_dependencies:
@@ -184,7 +184,7 @@ steps:
184 commands:
185 - pytest -s -v compile/fullgraph/test_full_cudagraph.py
186
187 -- label: Pytorch Nightly Dependency Override Check # 2min
187 +- label: ":nvidia: (H200) PyTorch Nightly Dependency Override Check"
188 key: pytorch-nightly-dependency-override-check
189 # if this test fails, it means the nightly torch version is not compatible with some
190 # of the dependencies. Please check the error message and add the package to whitelist
@@ -197,6 +197,7 @@ steps:
197 - bash standalone_tests/pytorch_nightly_dependency.sh
198 mirror:
199 amd:
200 + label: ":amd: (MI300) PyTorch Nightly Dependency Override Check"
201 dind: false
202 device: mi300_1
203 timeout_in_minutes: 30
.buildkite/test_areas/quantization.yaml
+4 -4
@@ -2,7 +2,7 @@ group: Quantization
2 depends_on:
3 - image-build
4 steps:
5 -- label: Quantization %N
5 +- label: ":nvidia: (H200) Quantization Shard %N"
6 device: h200_35gb
7 key: quantization
8 timeout_in_minutes: 40
@@ -21,7 +21,7 @@ steps:
21 # parameter. It was not exercised by the previous L4 job.
22 - VLLM_TEST_FORCE_LOAD_FORMAT=auto pytest -v -s quantization/ --ignore quantization/test_blackwell_moe.py -k 'not test_compressed_tensors_w4a8_fp8' --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
23
24 -- label: Quantized Fusions
24 +- label: ":nvidia: (H200) Quantized Fusions"
25 device: h200_35gb
26 key: quantized-fusions
27 timeout_in_minutes: 20
@@ -34,7 +34,7 @@ steps:
34 commands:
35 - pytest -v -s fusion/
36
37 -- label: Quantized MoE Test (B200)
37 +- label: ":nvidia: (B200) Quantized MoE"
38 key: quantized-moe-test-b200
39 timeout_in_minutes: 120
40 working_dir: "/vllm-workspace/"
@@ -52,7 +52,7 @@ steps:
52 commands:
53 - pytest -s -v tests/quantization/test_blackwell_moe.py
54
55 -- label: Quantized Models Test
55 +- label: ":nvidia: (H200) Quantized Models"
56 device: h200_35gb
57 key: quantized-models-test
58 timeout_in_minutes: 65
.buildkite/test_areas/ray_compat.yaml
+1 -1
@@ -2,7 +2,7 @@ group: Ray Compatibility
2 depends_on:
3 - image-build
4 steps:
5 -- label: Ray Dependency Compatibility Check
5 +- label: ":nvidia: (H200) Ray Dependency Compatibility Check"
6 key: ray-dependency-compatibility-check
7 # Informational only — does not block the pipeline.
8 # If this fails, it means the PR introduces a dependency that
.buildkite/test_areas/rust_frontend.yaml
+5 -5
@@ -2,7 +2,7 @@ group: Rust Frontend E2E
2 depends_on:
3 - image-build
4 steps:
5 -- label: Rust Frontend OpenAI Coverage
5 +- label: ":nvidia: (H200) Rust Frontend OpenAI Coverage"
6 timeout_in_minutes: 30
7 device: h200_18gb
8 working_dir: "/vllm-workspace/tests"
@@ -37,7 +37,7 @@ steps:
37 - pytest -v -s entrypoints/openai/test_uds.py
38 - pytest -v -s v1/sample/test_logprobs_e2e.py -k "test_prompt_logprobs_e2e_server"
39
40 -- label: Rust Frontend Serve/Admin Coverage
40 +- label: ":nvidia: (H200) Rust Frontend Serve/Admin Coverage"
41 timeout_in_minutes: 25
42 device: h200_18gb
43 working_dir: "/vllm-workspace/tests"
@@ -66,7 +66,7 @@ steps:
66 # /tokenizer_info is not implemented in the Rust frontend (the CLI flag is accepted as a no-op).
67 - pytest -v -s entrypoints/serve/tokenize/test_tokenization.py -k "not tokenizer_info"
68
69 -- label: Rust Frontend Core Correctness
69 +- label: ":nvidia: (H200) Rust Frontend Core Correctness"
70 timeout_in_minutes: 20
71 device: h200_18gb
72 working_dir: "/vllm-workspace/tests"
@@ -80,7 +80,7 @@ steps:
80 - export VLLM_WORKER_MULTIPROC_METHOD=spawn
81 - pytest -s entrypoints/openai/correctness/test_lmeval.py::test_lm_eval_accuracy_v1_engine
82
83 -- label: Rust Frontend Tool Use
83 +- label: ":nvidia: (H200) Rust Frontend Tool Use"
84 device: h200_35gb
85 timeout_in_minutes: 25
86 working_dir: "/vllm-workspace/tests"
@@ -95,7 +95,7 @@ steps:
95 - export VLLM_WORKER_MULTIPROC_METHOD=spawn
96 - pytest -v -s tool_use --ignore=tool_use/mistral --models llama3.2 -k "not test_response_format_with_tool_choice_required and not test_parallel_tool_calls_false and not test_tool_call_and_choice"
97
98 -- label: Rust Frontend Distributed
98 +- label: ":nvidia: (L4) Rust Frontend Distributed"
99 timeout_in_minutes: 25
100 num_devices: 4
101 working_dir: "/vllm-workspace/tests"
.buildkite/test_areas/rust_frontend_cargo.yaml
+2 -2
@@ -1,7 +1,7 @@
1 group: Rust Frontend Cargo
2 depends_on: []
3 steps:
4 -- label: Rust Frontend Cargo Style + Clippy
4 +- label: ":computer: (CPU) Rust Frontend Cargo Style + Clippy"
5 key: rust-frontend-cargo-style-clippy
6 depends_on: []
7 timeout_in_minutes: 20
@@ -15,7 +15,7 @@ steps:
15 commands:
16 - .buildkite/scripts/run-rust-frontend-cargo-ci.sh style-clippy
17
18 -- label: Rust Frontend Cargo Tests
18 +- label: ":computer: (CPU) Rust Frontend Cargo"
19 key: rust-frontend-cargo-tests
20 depends_on: []
21 timeout_in_minutes: 20
.buildkite/test_areas/samplers.yaml
+2 -1
@@ -2,7 +2,7 @@ group: Samplers
2 depends_on:
3 - image-build
4 steps:
5 -- label: Samplers Test
5 +- label: ":nvidia: (H200) Samplers"
6 device: h200_35gb
7 key: samplers-test
8 timeout_in_minutes: 40
@@ -19,6 +19,7 @@ steps:
19 - VLLM_USE_FLASHINFER_SAMPLER=1 pytest -v -s samplers
20 mirror:
21 amd:
22 + label: ":amd: (MI250) Samplers"
23 device: mi250_1
24 timeout_in_minutes: 40
25 depends_on:
.buildkite/test_areas/spec_decode.yaml
+23 -14
@@ -2,7 +2,7 @@ group: Spec Decode
2 depends_on:
3 - image-build
4 steps:
5 -- label: V1 Spec Decode
5 +- label: ":nvidia: (H200) V1 Spec Decode"
6 device: h200_35gb
7 key: v1-spec-decode
8 timeout_in_minutes: 40
@@ -23,13 +23,14 @@ steps:
23 - pytest -v -s -m 'not slow_test' v1/spec_decode
24 mirror:
25 amd:
26 + label: ":amd: (MI300) V1 Spec Decode"
27 dind: false
28 device: mi300_1
29 timeout_in_minutes: 50
30 depends_on:
31 - image-build-amd
32
32 -- label: "Spec Decode Eagle 1: DeepSeek + Qwen"
33 +- label: ":nvidia: (H200) Spec Decode Eagle 1: DeepSeek + Qwen"
34 key: spec-decode-eagle-1-deepseek-qwen
35 timeout_in_minutes: 25
36 device: h200_35gb
@@ -41,6 +42,7 @@ steps:
42 - pytest -v -s v1/e2e/spec_decode/eagle/ -k "deepseek_eagle or qwen3_eagle3"
43 mirror:
44 amd:
45 + label: ":amd: (MI300) Spec Decode Eagle 1: DeepSeek + Qwen"
46 dind: false
47 device: mi300_1
48 timeout_in_minutes: 25
@@ -55,7 +57,7 @@ steps:
57 - tests/v1/e2e/spec_decode/
58 - vllm/platforms/rocm.py
59
58 -- label: "Spec Decode Eagle 2: Llama 3 + Qwen VL + Other"
60 +- label: ":nvidia: (H200) Spec Decode Eagle 2: Llama 3 + Qwen VL + Other"
61 key: spec-decode-eagle-2-llama3-qwen-vl-other
62 timeout_in_minutes: 25
63 device: h200_35gb
@@ -67,6 +69,7 @@ steps:
69 - pytest -v -s v1/e2e/spec_decode/eagle/ -k "not deepseek_eagle and not qwen3_eagle3"
70 mirror:
71 amd:
72 + label: ":amd: (MI300) Spec Decode Eagle 2: Llama 3 + Qwen VL + Other"
73 dind: false
74 device: mi300_1
75 timeout_in_minutes: 25
@@ -81,7 +84,7 @@ steps:
84 - tests/v1/e2e/spec_decode/
85 - vllm/platforms/rocm.py
86
84 -- label: Spec Decode Eagle Nightly (B200)
87 +- label: ":nvidia: (B200) Spec Decode Eagle Nightly"
88 key: spec-decode-eagle-nightly-b200
89 timeout_in_minutes: 25
90 device: b200-k8s
@@ -93,7 +96,7 @@ steps:
96 commands:
97 - pytest -v -s v1/e2e/spec_decode/eagle/
98
96 -- label: Spec Decode Speculators + MTP
99 +- label: ":nvidia: (H200) Spec Decode Speculators + MTP"
100 key: spec-decode-speculators-mtp
101 timeout_in_minutes: 50
102 device: h200_35gb
@@ -108,6 +111,7 @@ steps:
111 - pytest -v -s v1/e2e/spec_decode/mtp/
112 mirror:
113 amd:
114 + label: ":amd: (MI300) Spec Decode Speculators + MTP"
115 dind: false
116 device: mi300_1
117 timeout_in_minutes: 75
@@ -123,7 +127,7 @@ steps:
127 - tests/v1/e2e/spec_decode/
128 - vllm/platforms/rocm.py
129
126 -- label: Spec Decode Speculators + MTP Nightly (B200)
130 +- label: ":nvidia: (B200) Spec Decode Speculators + MTP Nightly"
131 key: spec-decode-speculators-mtp-nightly-b200
132 timeout_in_minutes: 30
133 device: b200-k8s
@@ -137,7 +141,7 @@ steps:
141 - pytest -v -s v1/e2e/spec_decode/speculators/
142 - pytest -v -s v1/e2e/spec_decode/mtp/
143
140 -- label: Spec Decode Ngram + Suffix
144 +- label: ":nvidia: (H200) Spec Decode N-Gram + Suffix"
145 key: spec-decode-ngram-suffix
146 timeout_in_minutes: 20
147 device: h200_35gb
@@ -151,6 +155,7 @@ steps:
155 - python3 spec_decode/test_custom_proposer.py
156 mirror:
157 amd:
158 + label: ":amd: (MI300) Spec Decode N-Gram + Suffix"
159 dind: false
160 device: mi300_1
161 timeout_in_minutes: 35
@@ -165,7 +170,7 @@ steps:
170 - tests/v1/e2e/spec_decode/
171 - vllm/platforms/rocm.py
172
168 -- label: Spec Decode Draft Model
173 +- label: ":nvidia: (H200) Spec Decode Draft Model"
174 key: spec-decode-draft-model
175 timeout_in_minutes: 45
176 device: h200_18gb
@@ -177,6 +182,7 @@ steps:
182 - pytest -v -s v1/e2e/spec_decode/draft_model/
183 mirror:
184 amd:
185 + label: ":amd: (MI300) Spec Decode Draft Model"
186 dind: false
187 device: mi300_1
188 timeout_in_minutes: 55
@@ -191,7 +197,7 @@ steps:
197 - tests/v1/e2e/spec_decode/
198 - vllm/platforms/rocm.py
199
194 -- label: Spec Decode Draft Model Nightly (B200)
200 +- label: ":nvidia: (B200) Spec Decode Draft Model Nightly"
201 key: spec-decode-draft-model-nightly-b200
202 timeout_in_minutes: 40
203 device: b200-k8s
@@ -203,7 +209,7 @@ steps:
209 commands:
210 - pytest -v -s v1/e2e/spec_decode/draft_model/
211
206 -- label: Speculators Correctness Nightly (H100)
212 +- label: ":nvidia: (H100) Speculators Correctness Nightly"
213 key: speculators-correctness
214 timeout_in_minutes: 30
215 device: h100
@@ -218,7 +224,7 @@ steps:
224 - export VLLM_ALLOW_INSECURE_SERIALIZATION=1
225 - pytest -v -s v1/spec_decode/test_speculators_correctness.py -m slow_test
226
221 -- label: Spec Decode DeepSeek MTP Parallel Load (2xB200-2xMI300)
227 +- label: ":nvidia: (B200) Spec Decode DeepSeek MTP Parallel Load"
228 key: spec-decode-deepseek-mtp-parallel-load-2xb200-2xmi300
229 timeout_in_minutes: 30
230 device: b200-k8s
@@ -235,6 +241,7 @@ steps:
241 - pytest -v -s v1/e2e/spec_decode/test_mtp_parallel_load.py
242 mirror:
243 amd:
244 + label: ":amd: (MI300) Spec Decode DeepSeek MTP Parallel Load"
245 dind: false
246 device: mi300_2
247 timeout_in_minutes: 45
@@ -254,7 +261,7 @@ steps:
261 - tests/v1/e2e/spec_decode/test_mtp_parallel_load.py
262 - vllm/platforms/rocm.py
263
257 -- label: Spec Decode AL DFlash Nightly
264 +- label: ":nvidia: (H200) Spec Decode AL DFlash Nightly"
265 key: spec-decode-dflash-nightly
266 timeout_in_minutes: 90
267 device: h200_35gb
@@ -269,6 +276,7 @@ steps:
276 - pytest -v -s v1/e2e/spec_decode/acceptance_rates/dflash/
277 mirror:
278 amd:
279 + label: ":amd: (MI300) Spec Decode AL DFlash Nightly"
280 dind: false
281 device: mi300_1
282 timeout_in_minutes: 120
@@ -289,7 +297,7 @@ steps:
297 - tests/v1/e2e/spec_decode/
298 - vllm/platforms/rocm.py
299
292 -- label: Spec Decode AL DSpark Nightly
300 +- label: ":nvidia: (H200) Spec Decode AL DSpark Nightly"
301 key: spec-decode-dspark-nightly
302 timeout_in_minutes: 60
303 device: h200_35gb
@@ -304,6 +312,7 @@ steps:
312 - pytest -v -s v1/e2e/spec_decode/acceptance_rates/dspark/
313 mirror:
314 amd:
315 + label: ":amd: (MI300) Spec Decode AL DSpark Nightly"
316 dind: false
317 device: mi300_1
318 timeout_in_minutes: 90
@@ -328,7 +337,7 @@ steps:
337 - tests/v1/e2e/spec_decode/
338 - vllm/platforms/rocm.py
339
331 -- label: Spec Decode AL MTP + Other Acceptance Nightly
340 +- label: ":nvidia: (H200) Spec Decode AL MTP + Other Acceptance Nightly"
341 key: spec-decode-mtp-other-acceptance-nightly
342 timeout_in_minutes: 60
343 device: h200_35gb
.buildkite/test_areas/torch_abi.yaml
+1 -1
@@ -2,7 +2,7 @@ group: Torch ABI
2 depends_on:
3 - image-build
4 steps:
5 -- label: Torch Stable ABI Audit
5 +- label: ":nvidia: (L4) Torch Stable ABI Audit"
6 key: torch-stable-abi-audit
7 timeout_in_minutes: 5
8 source_file_dependencies:
.buildkite/test_areas/weight_loading.yaml
+2 -1
@@ -2,7 +2,7 @@ group: Weight Loading
2 depends_on:
3 - image-build
4 steps:
5 -- label: Weight Loading Multiple GPU # 33min
5 +- label: ":nvidia: (L4) Weight Loading Multi-GPU"
6 key: weight-loading-multiple-gpu
7 timeout_in_minutes: 50
8 working_dir: "/vllm-workspace/tests"
@@ -16,6 +16,7 @@ steps:
16 - bash weight_loading/run_model_weight_loading_test.sh -c weight_loading/models.txt
17 mirror:
18 amd:
19 + label: ":amd: (MI300) Weight Loading Multi-GPU"
20 dind: false
21 device: mi300_2
22 timeout_in_minutes: 35