@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / 81939e7733
Fork 0

A high-throughput and memory-efficient inference and serving engine for LLMs

Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git

Synced Nov 6, 23:00 UTC

[ROCm][CI] Making some tests optional to reduce workload (#36090)

Signed-off-by: Andreas Karatzas <akaratza@amd.com>

Andreas Karatzas committed Mar 10, 2026 at 18:45 UTC 81939e7733642f583d1731e5c9ef69dcd457b5e5
5 files changed +117 -34
.buildkite/test-amd.yaml
+117 -4
@@ -42,6 +42,7 @@ steps:
42 mirror_hardwares: [amdexperimental, amdproduction, amdtentative]
43 agent_pool: mi325_1
44 grade: Blocking
45 + optional: true
46 soft_fail: true
47 source_file_dependencies:
48 - requirements/nightly_torch_test.txt
@@ -67,6 +68,7 @@ steps:
68 timeout_in_minutes: 30
69 mirror_hardwares: [amdexperimental, amdproduction, amdtentative]
70 agent_pool: mi325_1
71 + optional: true
72 # grade: Blocking
73 source_file_dependencies:
74 - vllm/
@@ -97,6 +99,7 @@ steps:
99 timeout_in_minutes: 20
100 mirror_hardwares: [amdexperimental]
101 agent_pool: mi325_1
102 + optional: true
103 # grade: Blocking
104 source_file_dependencies:
105 - tests/standalone_tests/python_only_compile.sh
@@ -140,6 +143,7 @@ steps:
143 timeout_in_minutes: 40
144 mirror_hardwares: [amdexperimental, amdproduction]
145 agent_pool: mi325_1
146 + optional: true
147 # grade: Blocking
148 working_dir: "/vllm-workspace/tests"
149 fast_check: true
@@ -503,6 +507,7 @@ steps:
507 mirror_hardwares: [amdexperimental, amdproduction, amdtentative]
508 agent_pool: mi325_1
509 grade: Blocking
510 + optional: true
511 source_file_dependencies:
512 - vllm/
513 - tests/v1
@@ -520,6 +525,7 @@ steps:
525 timeout_in_minutes: 45
526 mirror_hardwares: [amdexperimental, amdproduction]
527 agent_pool: mi325_1
528 + optional: true
529 # grade: Blocking
530 working_dir: "/vllm-workspace/examples"
531 source_file_dependencies:
@@ -823,6 +829,7 @@ steps:
829 timeout_in_minutes: 90
830 mirror_hardwares: [amdexperimental, amdproduction]
831 agent_pool: mi325_1
832 + optional: true
833 # grade: Blocking
834 source_file_dependencies:
835 - csrc/
@@ -936,6 +943,7 @@ steps:
943 timeout_in_minutes: 25
944 mirror_hardwares: [amdexperimental, amdproduction]
945 agent_pool: mi325_1
946 + optional: true
947 # grade: Blocking
948 torch_nightly: true
949 source_file_dependencies:
@@ -1046,6 +1054,7 @@ steps:
1054 timeout_in_minutes: 60
1055 mirror_hardwares: [amdexperimental, amdproduction]
1056 agent_pool: mi325_1
1057 + optional: true
1058 source_file_dependencies:
1059 - vllm/
1060 - tests/models/multimodal
@@ -1059,6 +1068,7 @@ steps:
1068 timeout_in_minutes: 60
1069 mirror_hardwares: [amdexperimental, amdproduction]
1070 agent_pool: mi325_1
1071 + optional: true
1072 # grade: Blocking
1073 source_file_dependencies:
1074 - vllm/
@@ -1072,6 +1082,7 @@ steps:
1082 timeout_in_minutes: 100
1083 mirror_hardwares: [amdexperimental, amdproduction]
1084 agent_pool: mi325_1
1085 + optional: true
1086 # grade: Blocking
1087 torch_nightly: true
1088 source_file_dependencies:
@@ -1090,6 +1101,7 @@ steps:
1101 timeout_in_minutes: 10
1102 mirror_hardwares: [amdexperimental, amdproduction]
1103 agent_pool: mi325_1
1104 + optional: true
1105 # grade: Blocking
1106 working_dir: "/vllm-workspace/.buildkite/lm-eval-harness"
1107 source_file_dependencies:
@@ -1355,6 +1367,7 @@ steps:
1367 timeout_in_minutes: 60
1368 mirror_hardwares: [amdexperimental, amdproduction]
1369 agent_pool: mi325_2
1370 + optional: true
1371 # grade: Blocking
1372 working_dir: "/vllm-workspace/tests"
1373 num_gpus: 2
@@ -1393,6 +1406,7 @@ steps:
1406 timeout_in_minutes: 60
1407 mirror_hardwares: [amdexperimental, amdproduction]
1408 agent_pool: mi325_4
1409 + optional: true
1410 # grade: Blocking
1411 working_dir: "/vllm-workspace/tests"
1412 num_gpus: 4
@@ -1410,6 +1424,7 @@ steps:
1424 timeout_in_minutes: 30
1425 mirror_hardwares: [amdexperimental, amdproduction]
1426 agent_pool: mi325_4
1427 + optional: true
1428 # grade: Blocking
1429 num_gpus: 4
1430 source_file_dependencies:
@@ -1461,6 +1476,7 @@ steps:
1476 - label: NixlConnector PD accuracy tests (Distributed) # 30min
1477 mirror_hardwares: [amdexperimental, amdproduction]
1478 agent_pool: mi325_4
1479 + optional: true
1480 # grade: Blocking
1481 timeout_in_minutes: 30
1482 working_dir: "/vllm-workspace/tests"
@@ -1475,6 +1491,7 @@ steps:
1491 - label: DP EP NixlConnector PD accuracy tests (Distributed) # 15min
1492 mirror_hardwares: [amdexperimental, amdproduction]
1493 agent_pool: mi325_4
1494 + optional: true
1495 # grade: Blocking
1496 timeout_in_minutes: 15
1497 working_dir: "/vllm-workspace/tests"
@@ -1779,6 +1796,7 @@ steps:
1796 # in /vllm/tools/pre_commit/generate_nightly_torch_test.py
1797 mirror_hardwares: [amdexperimental, amdproduction, amdtentative]
1798 agent_pool: mi355_1
1799 + optional: true
1800 soft_fail: true
1801 source_file_dependencies:
1802 - requirements/nightly_torch_test.txt
@@ -1789,6 +1807,7 @@ steps:
1807 timeout_in_minutes: 15
1808 mirror_hardwares: [amdexperimental, amdproduction, amdtentative]
1809 agent_pool: mi355_1
1810 + optional: true
1811 source_file_dependencies:
1812 - vllm/
1813 - tests/multimodal
@@ -1801,6 +1820,7 @@ steps:
1820 timeout_in_minutes: 30
1821 mirror_hardwares: [amdexperimental, amdproduction, amdtentative]
1822 agent_pool: mi355_1
1823 + optional: true
1824 source_file_dependencies:
1825 - vllm/
1826 - tests/test_inputs.py
@@ -1830,6 +1850,7 @@ steps:
1850 timeout_in_minutes: 20
1851 mirror_hardwares: [amdexperimental]
1852 agent_pool: mi355_1
1853 + optional: true
1854 source_file_dependencies:
1855 - tests/standalone_tests/python_only_compile.sh
1856 - setup.py
@@ -1840,6 +1861,7 @@ steps:
1861 timeout_in_minutes: 30
1862 mirror_hardwares: [amdexperimental, amdproduction]
1863 agent_pool: mi355_1
1864 + optional: true
1865 fast_check: true
1866 torch_nightly: true
1867 source_file_dependencies:
@@ -1870,6 +1892,7 @@ steps:
1892 timeout_in_minutes: 40
1893 mirror_hardwares: [amdexperimental, amdproduction]
1894 agent_pool: mi355_1
1895 + optional: true
1896 working_dir: "/vllm-workspace/tests"
1897 fast_check: true
1898 torch_nightly: true
@@ -1887,6 +1910,7 @@ steps:
1910 timeout_in_minutes: 130
1911 mirror_hardwares: [amdexperimental]
1912 agent_pool: mi355_1
1913 + optional: true
1914 working_dir: "/vllm-workspace/tests"
1915 fast_check: true
1916 torch_nightly: true
@@ -1903,6 +1927,7 @@ steps:
1927 timeout_in_minutes: 50
1928 mirror_hardwares: [amdexperimental]
1929 agent_pool: mi355_1
1930 + optional: true
1931 working_dir: "/vllm-workspace/tests"
1932 fast_check: true
1933 torch_nightly: true
@@ -1921,6 +1946,7 @@ steps:
1946 timeout_in_minutes: 50
1947 mirror_hardwares: [amdexperimental]
1948 agent_pool: mi355_1
1949 + optional: true
1950 working_dir: "/vllm-workspace/tests"
1951 fast_check: true
1952 torch_nightly: true
@@ -1935,6 +1961,7 @@ steps:
1961 timeout_in_minutes: 50
1962 mirror_hardwares: [amdexperimental]
1963 agent_pool: mi355_1
1964 + optional: true
1965 working_dir: "/vllm-workspace/tests"
1966 fast_check: true
1967 torch_nightly: true
@@ -2013,6 +2040,7 @@ steps:
2040 timeout_in_minutes: 10
2041 mirror_hardwares: [amdexperimental]
2042 agent_pool: mi355_8
2043 + optional: true
2044 gpu: h100
2045 num_gpus: 8
2046 working_dir: "/vllm-workspace/tests"
@@ -2033,6 +2061,7 @@ steps:
2061 - label: EPLB Algorithm Test # 5min
2062 mirror_hardwares: [amdexperimental, amdproduction, amdtentative]
2063 agent_pool: mi355_1
2064 + optional: true
2065 timeout_in_minutes: 15
2066 working_dir: "/vllm-workspace/tests"
2067 source_file_dependencies:
@@ -2044,6 +2073,7 @@ steps:
2073 - label: EPLB Execution Test # 10min
2074 mirror_hardwares: [amdexperimental, amdproduction]
2075 agent_pool: mi355_4
2076 + optional: true
2077 timeout_in_minutes: 20
2078 working_dir: "/vllm-workspace/tests"
2079 num_gpus: 4
@@ -2058,6 +2088,7 @@ steps:
2088 timeout_in_minutes: 20
2089 mirror_hardwares: [amdexperimental, amdproduction]
2090 agent_pool: mi355_2
2091 + optional: true
2092 num_gpus: 2
2093 source_file_dependencies:
2094 - vllm/
@@ -2099,12 +2130,13 @@ steps:
2130 commands:
2131 - pytest -v -s engine test_sequence.py test_config.py test_logger.py test_vllm_port.py
2132
2133 +
2134 - label: V1 Test e2e + engine # 65min
2135 timeout_in_minutes: 90
2136 mirror_hardwares: [amdexperimental]
2105 - # The test uses 4 GPUs, but we schedule it on 8-GPU machines for stability.
2106 - # See discussion here: https://github.com/vllm-project/vllm/pull/31040
2107 - agent_pool: mi355_8
2137 + agent_pool: mi355_1
2138 + optional: true
2139 + # grade: Blocking
2140 source_file_dependencies:
2141 - vllm/
2142 - tests/v1
@@ -2114,10 +2146,39 @@ steps:
2146 - pytest -v -s v1/e2e
2147 - pytest -v -s v1/engine
2148
2149 +- label: V1 Test e2e (2 GPUs) # 65min
2150 + timeout_in_minutes: 90
2151 + mirror_hardwares: [amdexperimental]
2152 + agent_pool: mi355_2
2153 + optional: true
2154 + # grade: Blocking
2155 + source_file_dependencies:
2156 + - vllm/
2157 + - tests/v1
2158 + commands:
2159 + # Only run tests that need exactly 2 GPUs
2160 + - pytest -v -s v1/e2e/test_spec_decode.py -k "tensor_parallelism"
2161 +
2162 +- label: V1 Test e2e (4 GPUs) # 65min
2163 + timeout_in_minutes: 90
2164 + mirror_hardwares: [amdexperimental]
2165 + # The test uses 4 GPUs, but we schedule it on 8-GPU machines for stability.
2166 + # See discussion here: https://github.com/vllm-project/vllm/pull/31040
2167 + agent_pool: mi355_4
2168 + optional: true
2169 + # grade: Blocking
2170 + source_file_dependencies:
2171 + - vllm/
2172 + - tests/v1
2173 + commands:
2174 + # Only run tests that need 4 GPUs
2175 + - pytest -v -s v1/e2e/test_spec_decode.py -k "eagle_correctness_heavy"
2176 +
2177 - label: V1 Test entrypoints # 35min
2178 timeout_in_minutes: 50
2179 mirror_hardwares: [amdexperimental, amdproduction, amdtentative]
2180 agent_pool: mi355_1
2181 + optional: true
2182 source_file_dependencies:
2183 - vllm/
2184 - tests/v1
@@ -2128,6 +2189,7 @@ steps:
2189 timeout_in_minutes: 60
2190 mirror_hardwares: [amdexperimental]
2191 agent_pool: mi355_1
2192 + optional: true
2193 source_file_dependencies:
2194 - vllm/
2195 - tests/v1
@@ -2150,7 +2212,19 @@ steps:
2212 - pip install -U git+https://github.com/robertgshaw2-redhat/lm-evaluation-harness.git@streaming-api
2213 - pytest -v -s entrypoints/openai/correctness/test_lmeval.py::test_lm_eval_accuracy_v1_engine
2214
2153 -# TODO: Add the "V1 Test attention (MI300)" test group
2215 +- label: V1 Test attention (H100) # 10min
2216 + mirror_hardwares: [amdexperimental]
2217 + agent_pool: mi355_1
2218 + optional: true
2219 + timeout_in_minutes: 30
2220 + gpu: h100
2221 + source_file_dependencies:
2222 + - vllm/config/attention.py
2223 + - vllm/model_executor/layers/attention
2224 + - vllm/v1/attention
2225 + - tests/v1/attention
2226 + commands:
2227 + - pytest -v -s v1/attention
2228
2229 - label: Batch Invariance Tests (H100) # 10min
2230 mirror_hardwares: [amdexperimental]
@@ -2200,6 +2274,7 @@ steps:
2274 timeout_in_minutes: 45
2275 mirror_hardwares: [amdexperimental, amdproduction]
2276 agent_pool: mi355_1
2277 + optional: true
2278 working_dir: "/vllm-workspace/examples"
2279 source_file_dependencies:
2280 - vllm/entrypoints
@@ -2234,6 +2309,7 @@ steps:
2309 timeout_in_minutes: 15
2310 mirror_hardwares: [amdexperimental, amdproduction]
2311 agent_pool: mi355_1
2312 + optional: true
2313 source_file_dependencies:
2314 - vllm/
2315 - tests/cuda
@@ -2245,6 +2321,7 @@ steps:
2321 timeout_in_minutes: 75
2322 mirror_hardwares: [amdexperimental]
2323 agent_pool: mi355_1
2324 + optional: true
2325 source_file_dependencies:
2326 - vllm/model_executor/layers
2327 - vllm/sampling_metadata.py
@@ -2277,6 +2354,7 @@ steps:
2354 timeout_in_minutes: 30
2355 mirror_hardwares: [amdexperimental, amdproduction]
2356 agent_pool: mi355_1
2357 + optional: true
2358 torch_nightly: true
2359 source_file_dependencies:
2360 - vllm/
@@ -2293,6 +2371,7 @@ steps:
2371 timeout_in_minutes: 30
2372 mirror_hardwares: [amdexperimental, amdproduction]
2373 agent_pool: mi355_1
2374 + optional: true
2375 torch_nightly: true
2376 source_file_dependencies:
2377 - vllm/
@@ -2308,6 +2387,7 @@ steps:
2387 timeout_in_minutes: 40
2388 mirror_hardwares: [amdexperimental, amdproduction]
2389 agent_pool: mi355_1
2390 + optional: true
2391 # grade: Blocking
2392 torch_nightly: true
2393 source_file_dependencies:
@@ -2325,6 +2405,7 @@ steps:
2405 timeout_in_minutes: 20
2406 mirror_hardwares: [amdexperimental, amdproduction]
2407 agent_pool: mi355_1
2408 + optional: true
2409 source_file_dependencies:
2410 - tests/v1/cudagraph
2411 - vllm/v1/cudagraph_dispatcher.py
@@ -2338,6 +2419,7 @@ steps:
2419 timeout_in_minutes: 75
2420 mirror_hardwares: [amdexperimental, amdproduction]
2421 agent_pool: mi355_1
2422 + optional: true
2423 source_file_dependencies:
2424 - csrc/
2425 - tests/kernels/core
@@ -2349,6 +2431,7 @@ steps:
2431 timeout_in_minutes: 35
2432 mirror_hardwares: [amdexperimental, amdproduction]
2433 agent_pool: mi355_1
2434 + optional: true
2435 source_file_dependencies:
2436 - csrc/attention/
2437 - vllm/v1/attention
@@ -2363,6 +2446,7 @@ steps:
2446 timeout_in_minutes: 90
2447 mirror_hardwares: [amdexperimental]
2448 agent_pool: mi355_1
2449 + optional: true
2450 source_file_dependencies:
2451 - csrc/quantization/
2452 - vllm/model_executor/layers/quantization
@@ -2375,6 +2459,7 @@ steps:
2459 timeout_in_minutes: 60
2460 mirror_hardwares: [amdexperimental, amdproduction]
2461 agent_pool: mi355_1
2462 + optional: true
2463 source_file_dependencies:
2464 - csrc/quantization/cutlass_w8a8/moe/
2465 - csrc/moe/
@@ -2391,6 +2476,7 @@ steps:
2476 timeout_in_minutes: 45
2477 mirror_hardwares: [amdexperimental, amdproduction]
2478 agent_pool: mi355_1
2479 + optional: true
2480 source_file_dependencies:
2481 - csrc/mamba/
2482 - tests/kernels/mamba
@@ -2422,6 +2508,7 @@ steps:
2508 timeout_in_minutes: 30
2509 mirror_hardwares: [amdexperimental, amdproduction]
2510 agent_pool: mi355_1
2511 + optional: true
2512 source_file_dependencies:
2513 - vllm/utils/import_utils.py
2514 - tests/kernels/helion/
@@ -2434,6 +2521,7 @@ steps:
2521 torch_nightly: true
2522 mirror_hardwares: [amdexperimental, amdproduction]
2523 agent_pool: mi355_1
2524 + optional: true
2525 source_file_dependencies:
2526 - vllm/engine/arg_utils.py
2527 - vllm/config/model.py
@@ -2450,6 +2538,7 @@ steps:
2538 timeout_in_minutes: 20
2539 mirror_hardwares: [amdexperimental, amdproduction]
2540 agent_pool: mi355_1
2541 + optional: true
2542 working_dir: "/vllm-workspace/.buildkite"
2543 source_file_dependencies:
2544 - benchmarks/
@@ -2460,6 +2549,7 @@ steps:
2549 timeout_in_minutes: 20
2550 mirror_hardwares: [amdexperimental, amdproduction]
2551 agent_pool: mi355_1
2552 + optional: true
2553 source_file_dependencies:
2554 - vllm/
2555 - tests/benchmarks/
@@ -2470,6 +2560,7 @@ steps:
2560 timeout_in_minutes: 90
2561 mirror_hardwares: [amdexperimental, amdproduction]
2562 agent_pool: mi355_1
2563 + optional: true
2564 source_file_dependencies:
2565 - csrc/
2566 - vllm/model_executor/layers/quantization
@@ -2490,6 +2581,7 @@ steps:
2581 timeout_in_minutes: 75
2582 mirror_hardwares: [amdexperimental]
2583 agent_pool: mi355_1
2584 + optional: true
2585 source_file_dependencies:
2586 - csrc/
2587 - vllm/model_executor/layers/quantization
@@ -2501,6 +2593,7 @@ steps:
2593 timeout_in_minutes: 15
2594 mirror_hardwares: [amdexperimental, amdproduction]
2595 agent_pool: mi355_1
2596 + optional: true
2597 source_file_dependencies:
2598 - csrc/
2599 - vllm/entrypoints/openai/
@@ -2517,6 +2610,7 @@ steps:
2610 timeout_in_minutes: 45
2611 mirror_hardwares: [amdexperimental, amdproduction]
2612 agent_pool: mi355_1
2613 + optional: true
2614 torch_nightly: true
2615 source_file_dependencies:
2616 - vllm/
@@ -2529,6 +2623,7 @@ steps:
2623 timeout_in_minutes: 45
2624 mirror_hardwares: [amdexperimental, amdproduction]
2625 agent_pool: mi355_1
2626 + optional: true
2627 torch_nightly: true
2628 source_file_dependencies:
2629 - vllm/model_executor/models/
@@ -2548,6 +2643,7 @@ steps:
2643 timeout_in_minutes: 45
2644 mirror_hardwares: [amdexperimental]
2645 agent_pool: mi355_1
2646 + optional: true
2647 torch_nightly: true
2648 source_file_dependencies:
2649 - vllm/
@@ -2560,6 +2656,7 @@ steps:
2656 - label: Basic Models Test (Other CPU) # 5min
2657 mirror_hardwares: [amdexperimental, amdproduction]
2658 agent_pool: mi355_1
2659 + optional: true
2660 timeout_in_minutes: 10
2661 torch_nightly: true
2662 source_file_dependencies:
@@ -2574,6 +2671,7 @@ steps:
2671 timeout_in_minutes: 25
2672 mirror_hardwares: [amdexperimental, amdproduction]
2673 agent_pool: mi355_1
2674 + optional: true
2675 torch_nightly: true
2676 source_file_dependencies:
2677 - vllm/
@@ -2587,6 +2685,7 @@ steps:
2685 timeout_in_minutes: 45
2686 mirror_hardwares: [amdexperimental]
2687 agent_pool: mi355_1
2688 + optional: true
2689 torch_nightly: true
2690 source_file_dependencies:
2691 - vllm/model_executor/models/
@@ -2607,6 +2706,7 @@ steps:
2706 timeout_in_minutes: 75
2707 mirror_hardwares: [amdexperimental]
2708 agent_pool: mi355_1
2709 + optional: true
2710 torch_nightly: true
2711 source_file_dependencies:
2712 - vllm/
@@ -2676,6 +2776,7 @@ steps:
2776 timeout_in_minutes: 60
2777 mirror_hardwares: [amdexperimental]
2778 agent_pool: mi355_1
2779 + optional: true
2780 source_file_dependencies:
2781 - vllm/
2782 - tests/models/multimodal
@@ -2688,6 +2789,7 @@ steps:
2789 timeout_in_minutes: 60
2790 mirror_hardwares: [amdexperimental]
2791 agent_pool: mi355_1
2792 + optional: true
2793 source_file_dependencies:
2794 - vllm/
2795 - tests/models/multimodal
@@ -2699,6 +2801,7 @@ steps:
2801 timeout_in_minutes: 100
2802 mirror_hardwares: [amdexperimental]
2803 agent_pool: mi355_1
2804 + optional: true
2805 torch_nightly: true
2806 source_file_dependencies:
2807 - vllm/
@@ -2716,6 +2819,7 @@ steps:
2819 timeout_in_minutes: 10
2820 mirror_hardwares: [amdexperimental, amdproduction]
2821 agent_pool: mi355_1
2822 + optional: true
2823 working_dir: "/vllm-workspace/.buildkite/lm-eval-harness"
2824 source_file_dependencies:
2825 - vllm/multimodal/
@@ -2772,6 +2876,7 @@ steps:
2876 timeout_in_minutes: 60
2877 mirror_hardwares: [amdexperimental, amdproduction]
2878 agent_pool: mi355_1
2879 + optional: true
2880 source_file_dependencies:
2881 - vllm/model_executor/layers/quantization
2882 - tests/models/quantization
@@ -2923,6 +3028,7 @@ steps:
3028 timeout_in_minutes: 20
3029 mirror_hardwares: [amdexperimental, amdproduction]
3030 agent_pool: mi355_2
3031 + optional: true
3032 working_dir: "/vllm-workspace/tests"
3033 num_gpus: 2
3034 source_file_dependencies:
@@ -3005,6 +3111,7 @@ steps:
3111 timeout_in_minutes: 50
3112 mirror_hardwares: [amdexperimental]
3113 agent_pool: mi355_2
3114 + optional: true
3115 working_dir: "/vllm-workspace/tests"
3116 num_gpus: 2
3117 source_file_dependencies:
@@ -3026,6 +3133,7 @@ steps:
3133 timeout_in_minutes: 60
3134 mirror_hardwares: [amdexperimental, amdproduction]
3135 agent_pool: mi355_2
3136 + optional: true
3137 working_dir: "/vllm-workspace/tests"
3138 num_gpus: 2
3139 source_file_dependencies:
@@ -3063,6 +3171,7 @@ steps:
3171 timeout_in_minutes: 60
3172 mirror_hardwares: [amdexperimental, amdproduction]
3173 agent_pool: mi355_4
3174 + optional: true
3175 working_dir: "/vllm-workspace/tests"
3176 num_gpus: 4
3177 source_file_dependencies:
@@ -3079,6 +3188,7 @@ steps:
3188 timeout_in_minutes: 30
3189 mirror_hardwares: [amdexperimental, amdproduction]
3190 agent_pool: mi355_4
3191 + optional: true
3192 num_gpus: 4
3193 source_file_dependencies:
3194 - vllm/lora
@@ -3127,6 +3237,7 @@ steps:
3237 - label: NixlConnector PD accuracy tests (Distributed) # 30min
3238 mirror_hardwares: [amdexperimental, amdproduction]
3239 agent_pool: mi355_4
3240 + optional: true
3241 timeout_in_minutes: 30
3242 working_dir: "/vllm-workspace/tests"
3243 num_gpus: 4
@@ -3140,6 +3251,7 @@ steps:
3251 - label: DP EP NixlConnector PD accuracy tests (Distributed) # 15min
3252 mirror_hardwares: [amdexperimental, amdproduction]
3253 agent_pool: mi355_4
3254 + optional: true
3255 timeout_in_minutes: 15
3256 working_dir: "/vllm-workspace/tests"
3257 num_gpus: 4
@@ -3278,6 +3390,7 @@ steps:
3390 - label: ROCm LM Eval Large Models (8 Card)
3391 mirror_hardwares: [amdproduction]
3392 agent_pool: mi355_8
3393 + optional: true
3394 num_gpus: 8
3395 working_dir: "/vllm-workspace/.buildkite/lm-eval-harness"
3396 commands:
.buildkite/test_areas/basic_correctness.yaml
-5
@@ -14,8 +14,3 @@ steps:
14 - pytest -v -s basic_correctness/test_cumem.py
15 - pytest -v -s basic_correctness/test_basic_correctness.py
16 - pytest -v -s basic_correctness/test_cpu_offload.py
17 - mirror:
18 - amd:
19 - device: mi325_1
20 - depends_on:
21 - - image-build-amd
.buildkite/test_areas/entrypoints.yaml
-15
@@ -24,11 +24,6 @@ steps:
24 - pytest -v -s entrypoints/llm --ignore=entrypoints/llm/test_generate.py --ignore=entrypoints/llm/test_collective_rpc.py
25 - pytest -v -s entrypoints/llm/test_generate.py # it needs a clean process
26 - pytest -v -s entrypoints/offline_mode # Needs to avoid interference with other tests
27 - mirror:
28 - amd:
29 - device: mi325_1
30 - depends_on:
31 - - image-build-amd
27
28 - label: Entrypoints Integration (API Server 1)
29 timeout_in_minutes: 130
@@ -60,11 +55,6 @@ steps:
55 - pytest -v -s entrypoints/instrumentator
56 - PYTHONPATH=/vllm-workspace pytest -v -s entrypoints/rpc
57 - pytest -v -s tool_use
63 - mirror:
64 - amd:
65 - device: mi325_1
66 - depends_on:
67 - - image-build-amd
58
59 - label: Entrypoints Integration (Pooling)
60 timeout_in_minutes: 50
@@ -75,11 +65,6 @@ steps:
65 commands:
66 - export VLLM_WORKER_MULTIPROC_METHOD=spawn
67 - pytest -v -s entrypoints/pooling
78 - mirror:
79 - amd:
80 - device: mi325_1
81 - depends_on:
82 - - image-build-amd
68
69 - label: Entrypoints Integration (Responses API)
70 timeout_in_minutes: 50
.buildkite/test_areas/misc.yaml
-5
@@ -88,11 +88,6 @@ steps:
88 - python3 offline_inference/spec_decode.py --test --method eagle --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 2048
89 # https://github.com/vllm-project/vllm/pull/26682 uses slightly more memory in PyTorch 2.9+ causing this test to OOM in 1xL4 GPU
90 - python3 offline_inference/spec_decode.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
91 - mirror:
92 - amd:
93 - device: mi325_1
94 - depends_on:
95 - - image-build-amd
91
92 - label: Metrics, Tracing (2 GPUs)
93 timeout_in_minutes: 20
.buildkite/test_areas/plugins.yaml
-5
@@ -39,8 +39,3 @@ steps:
39 - pytest -v -s entrypoints/openai/test_oot_registration.py # it needs a clean process
40 - pytest -v -s models/test_oot_registration.py # it needs a clean process
41 - pytest -v -s plugins/lora_resolvers # unit tests for in-tree lora resolver plugins
42 - mirror:
43 - amd:
44 - device: mi325_2
45 - depends_on:
46 - - image-build-amd