[ROCm][CI] Expand single-GPU coverage on MI355 DPX (#57599)
Signed-off-by: Sheral Kumar <shekumar@amd.com> Signed-off-by: Andreas Karatzas <Andreas.Karatzas@amd.com> Signed-off-by: Andreas Karatzas <akaratza@amd.com> Signed-off-by: Andreas Karatzas <andreas.karatzas@protonmail.com> Co-authored-by: Andreas Karatzas <Andreas.Karatzas@amd.com> Co-authored-by: Codex <noreply@openai.com> Co-authored-by: Andreas Karatzas <akaratza@amd.com> Co-authored-by: OpenAI Codex <codex@openai.com> Co-authored-by: Andreas Karatzas <andreas.karatzas@protonmail.com>
Sheral Kumar committed
Sep 25, 2026 at 14:27 UTC
22bbe3f1023a68a7d1f2de566dd4242fdcfd36c3
25 files changed
+1314
-1232
.buildkite/test-amd.yaml
+1078
-1078
@@ -538,25 +538,6 @@ steps:
538
539
#----------------------------------------------------- mi300 · basic_correctness -----------------------------------------------------#
540
541
-- label: ":amd: (MI300) Basic Correctness" # TBD
542
- timeout_in_minutes: 60
543
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
544
- dind: false
545
- agent_pool: mi300_1
546
- fast_check: true
547
- working_dir: "/vllm-workspace/tests"
548
- source_file_dependencies:
549
- - vllm/
550
- - tests/basic_correctness/test_basic_correctness
551
- - tests/basic_correctness/test_cpu_offload
552
- - tests/basic_correctness/test_mem.py
553
- optional: true
554
- commands:
555
- - export VLLM_WORKER_MULTIPROC_METHOD=spawn
556
- - pytest -v -s basic_correctness/test_mem.py
557
- - VLLM_TARGET_TEST_SUITE=MI300 pytest -v -s basic_correctness/test_basic_correctness.py
558
- - pytest -v -s basic_correctness/test_cpu_offload.py
559
-
541
- label: ":amd: (MI300) Distributed Models Shard %N"
542
timeout_in_minutes: 80
543
parallelism: 3
@@ -589,22 +570,6 @@ steps:
570
commands:
571
- case "$$BUILDKITE_PARALLEL_JOB" in 0) TARGET_TEST_SUITE=MI300 pytest basic_correctness/ -v -s -m 'distributed(num_gpus=2)' && HIP_VISIBLE_DEVICES=0,1 pytest -v -s model_executor/model_loader/test_sharded_state_loader.py -m '(not slow_test)' ;; 1) pytest models/transformers/test_backend.py -v -s -m 'distributed(num_gpus=2)' && pytest models/language -v -s -m 'distributed(num_gpus=2)' && VLLM_ROCM_USE_AITER=1 pytest -v -s models/test_hyv4_rocm.py -m 'distributed(num_gpus=2)' ;; 2) pytest models/multimodal -v -s -m 'distributed(num_gpus=2)' --ignore models/multimodal/generation/test_whisper.py --ignore models/multimodal/generation/test_phi4siglip.py && pytest models/multimodal/generation/test_phi4siglip.py -v -s -m 'distributed(num_gpus=2)' && VLLM_WORKER_MULTIPROC_METHOD=spawn pytest models/multimodal/generation/test_whisper.py -v -s -m 'distributed(num_gpus=2)' && pytest -v -s models/test_vision.py -m 'distributed(num_gpus=2)' ;; *) exit 2 ;; esac
572
592
-#-------------------------------------------------------- mi300 · benchmarks ---------------------------------------------------------#
593
-
594
-- label: ":amd: (MI300) Benchmarks CLI" # TBD
595
- timeout_in_minutes: 40
596
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
597
- dind: false
598
- agent_pool: mi300_1
599
- optional: true
600
- working_dir: "/vllm-workspace/tests"
601
- source_file_dependencies:
602
- - vllm/
603
- - tests/benchmarks/
604
- - rust/src/bench/tests/python_serve_flags.txt
605
- commands:
606
- - pytest -v -s benchmarks/
607
-
573
#---------------------------------------------------------- mi300 · compile ----------------------------------------------------------#
574
575
- label: ":amd: (MI300) PyTorch Compilation" # TBD
@@ -922,22 +887,6 @@ steps:
887
- pytest -v -s distributed/test_shm_buffer.py
888
- pytest -v -s distributed/test_shm_storage.py
889
925
-- label: ":amd: (MI300) EPLB Algorithm" # TBD
926
- timeout_in_minutes: 30
927
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
928
- dind: false
929
- agent_pool: mi300_1
930
- optional: true
931
- working_dir: "/vllm-workspace/tests"
932
- source_file_dependencies:
933
- - vllm/distributed/eplb
934
- - tests/distributed/test_eplb_algo.py
935
- - tests/distributed/test_eplb_utils.py
936
- - vllm/platforms/rocm.py
937
- commands:
938
- - pytest -v -s distributed/test_eplb_algo.py
939
- - pytest -v -s distributed/test_eplb_utils.py
940
-
890
- label: ":amd: (MI300) EPLB Execution" # TBD
891
timeout_in_minutes: 35
892
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
@@ -1024,7 +973,7 @@ steps:
973
- pytest -v -s models/test_vision.py -m 'distributed(num_gpus=4)'
974
975
- label: ":amd: (MI300) Distributed Torchrun + Examples" # TBD
1027
- timeout_in_minutes: 35
976
+ timeout_in_minutes: 55
977
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
978
dind: false
979
agent_pool: mi300_4
@@ -1155,38 +1104,6 @@ steps:
1104
- pytest -v -s distributed/test_symm_mem_allreduce.py
1105
- pytest -v -s distributed/test_multiproc_executor.py::test_multiproc_executor_multi_node
1106
1158
-#---------------------------------------------------------- mi300 · engine -----------------------------------------------------------#
1159
-
1160
-- label: ":amd: (MI300) Engine Core" # TBD
1161
- timeout_in_minutes: 40
1162
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1163
- dind: false
1164
- agent_pool: mi300_1
1165
- optional: true
1166
- working_dir: "/vllm-workspace/tests"
1167
- source_file_dependencies:
1168
- - vllm/
1169
- - tests/engine
1170
- - tests/test_sequence
1171
- - tests/test_config
1172
- - tests/test_logger
1173
- - tests/test_vllm_port
1174
- - tests/jit_monitor/test_hooks.py
1175
- - tests/jit_monitor/test_hooks_gpu.py
1176
- - vllm/compilation/
1177
- - vllm/config/
1178
- - vllm/engine/
1179
- - vllm/entrypoints/logger.py
1180
- - vllm/envs.py
1181
- - vllm/logger.py
1182
- - vllm/logging_utils/
1183
- - vllm/platforms/
1184
- - vllm/sequence.py
1185
- - vllm/triton_utils/
1186
- - vllm/utils/
1187
- commands:
1188
- - 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
1189
-
1107
#-------------------------------------------------------- mi300 · entrypoints --------------------------------------------------------#
1108
1109
- label: ":amd: (MI300) Entrypoints Unit" # TBD
@@ -1208,95 +1125,6 @@ steps:
1125
- pytest -v -s entrypoints/weight_transfer
1126
- pytest -v -s entrypoints/launchers
1127
1211
-- label: ":amd: (MI300) Entrypoints Integration (LLM)" # TBD
1212
- timeout_in_minutes: 90
1213
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1214
- dind: false
1215
- agent_pool: mi300_1
1216
- optional: true
1217
- fast_check: true
1218
- working_dir: "/vllm-workspace/tests"
1219
- source_file_dependencies:
1220
- - vllm/
1221
- - tests/entrypoints/llm
1222
- commands:
1223
- - export VLLM_WORKER_MULTIPROC_METHOD=spawn
1224
- - pytest -v -s entrypoints/llm --ignore=entrypoints/llm/test_generate.py --ignore=entrypoints/llm/test_collective_rpc.py --ignore=entrypoints/llm/offline_mode
1225
- - pytest -v -s entrypoints/llm/test_generate.py # it needs a clean process
1226
- - pytest -v -s entrypoints/llm/offline_mode # Needs to avoid interference with other tests
1227
-
1228
-- label: ":amd: (MI300) Entrypoints Integration (API Server) %N"
1229
- timeout_in_minutes: 65
1230
- parallelism: 4
1231
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1232
- dind: false
1233
- agent_pool: mi300_1
1234
- optional: true
1235
- fast_check: true
1236
- working_dir: "/vllm-workspace/tests"
1237
- source_file_dependencies:
1238
- - vllm/
1239
- - tests/entrypoints/serve
1240
- - tests/entrypoints/scale_out
1241
- commands:
1242
- - export VLLM_WORKER_MULTIPROC_METHOD=spawn
1243
- - pytest -v -s entrypoints/serve --ignore=entrypoints/serve/dev/rpc --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
1244
- - if [ "$$BUILDKITE_PARALLEL_JOB" = "1" ]; then PYTHONPATH=/vllm-workspace pytest -v -s entrypoints/serve/dev/rpc; fi
1245
- - pytest -v -s entrypoints/scale_out --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
1246
-
1247
-- label: ":amd: (MI300) Entrypoints Integration (OpenAI API completion)" # TBD
1248
- timeout_in_minutes: 65
1249
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1250
- dind: false
1251
- agent_pool: mi300_1
1252
- fast_check: true
1253
- optional: true
1254
- working_dir: "/vllm-workspace/tests"
1255
- source_file_dependencies:
1256
- - vllm/
1257
- - tests/entrypoints/openai
1258
- commands:
1259
- - export VLLM_WORKER_MULTIPROC_METHOD=spawn
1260
- - pytest -v -s entrypoints/openai/completion --ignore=entrypoints/openai/completion/test_tensorizer_entrypoint.py
1261
- - pytest -v -s entrypoints/openai --ignore=entrypoints/openai/completion --ignore=entrypoints/openai/chat_completion --ignore=entrypoints/openai/responses --ignore=entrypoints/openai/correctness
1262
-
1263
-- label: ":amd: (MI300) Entrypoints Integration (OpenAI API chat_completion)" # TBD
1264
- timeout_in_minutes: 70
1265
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1266
- dind: false
1267
- agent_pool: mi300_1
1268
- fast_check: true
1269
- optional: true
1270
- working_dir: "/vllm-workspace/tests"
1271
- source_file_dependencies:
1272
- - vllm/
1273
- - tests/entrypoints/openai
1274
- commands:
1275
- - export VLLM_WORKER_MULTIPROC_METHOD=spawn
1276
- - pytest -v -s entrypoints/openai/chat_completion
1277
-
1278
-- label: ":amd: (MI300) Entrypoints Integration (API Server Generate)" # TBD
1279
- timeout_in_minutes: 65
1280
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1281
- dind: false
1282
- agent_pool: mi300_1
1283
- optional: true
1284
- fast_check: true
1285
- working_dir: "/vllm-workspace/tests"
1286
- source_file_dependencies:
1287
- - vllm/
1288
- - tests/tool_use
1289
- - tests/entrypoints/tool_parsers
1290
- - tests/entrypoints/anthropic
1291
- - tests/entrypoints/cohere
1292
- - tests/entrypoints/generate
1293
- commands:
1294
- - pytest -v -s tool_use
1295
- - pytest -v -s entrypoints/tool_parsers
1296
- - pytest -v -s entrypoints/generate
1297
- - pytest -v -s entrypoints/anthropic
1298
- - pytest -v -s entrypoints/cohere
1299
-
1128
- label: ":amd: (MI300) Entrypoints Integration (Responses API)" # TBD
1129
timeout_in_minutes: 50
1130
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
@@ -1341,26 +1169,6 @@ steps:
1169
- export VLLM_WORKER_MULTIPROC_METHOD=spawn
1170
- pytest -v -s entrypoints/multimodal
1171
1344
-- label: ":amd: (MI300) OpenAI API Correctness" # TBD
1345
- timeout_in_minutes: 30
1346
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1347
- dind: false
1348
- agent_pool: mi300_1
1349
- optional: true
1350
- working_dir: "/vllm-workspace/tests"
1351
- source_file_dependencies:
1352
- - csrc/
1353
- - vllm/entrypoints/openai/
1354
- - vllm/model_executor/layers/
1355
- - vllm/v1/attention/backends/
1356
- - vllm/v1/attention/selector.py
1357
- - vllm/_aiter_ops.py
1358
- - vllm/platforms/rocm.py
1359
- - vllm/model_executor/model_loader/
1360
- commands:
1361
- - bash ../tools/install_torchcodec_rocm.sh || exit 1
1362
- - pytest -s entrypoints/openai/correctness/
1363
-
1172
- label: ":amd: (MI300) Scale-out EC Connector E2E"
1173
timeout_in_minutes: 60
1174
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
@@ -1380,46 +1188,46 @@ steps:
1188
1189
#----------------------------------------------------------- mi300 · evals -----------------------------------------------------------#
1190
1383
-- label: ":amd: (MI300) DeepSeek V2-Lite Prefetch Offload Accuracy" # TBD
1384
- timeout_in_minutes: 30
1191
+- label: ":amd: (MI300) LM Eval Small Models" # TBD
1192
+ timeout_in_minutes: 45
1193
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1194
dind: false
1195
agent_pool: mi300_1
1388
- num_gpus: 1
1389
- working_dir: "/vllm-workspace"
1196
+ working_dir: "/vllm-workspace/tests"
1197
source_file_dependencies:
1198
+ - csrc/
1199
+ - vllm/model_executor/layers/quantization
1200
- vllm/model_executor/models/
1201
- vllm/model_executor/model_loader/
1393
- - vllm/model_executor/layers/fused_moe/
1394
- - vllm/model_executor/layers/quantization/
1202
- vllm/v1/attention/backends/
1396
- - vllm/v1/attention/backends/mla/
1203
- vllm/v1/attention/selector.py
1398
- - .buildkite/scripts/scheduled_integration_test/
1204
- vllm/_aiter_ops.py
1205
- vllm/platforms/rocm.py
1206
optional: true
1207
commands:
1403
- - bash .buildkite/scripts/scheduled_integration_test/deepseek_v2_lite_prefetch_offload.sh 0.25 200 8030
1208
+ - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-small.txt
1209
1405
-- label: ":amd: (MI300) LM Eval Small Models" # TBD
1406
- timeout_in_minutes: 45
1210
+- label: ":amd: (MI300) DeepSeek V2-Lite Prefetch Offload Accuracy" # TBD
1211
+ timeout_in_minutes: 30
1212
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1213
dind: false
1214
agent_pool: mi300_1
1410
- working_dir: "/vllm-workspace/tests"
1215
+ num_gpus: 1
1216
+ working_dir: "/vllm-workspace"
1217
source_file_dependencies:
1412
- - csrc/
1413
- - vllm/model_executor/layers/quantization
1218
- vllm/model_executor/models/
1219
- vllm/model_executor/model_loader/
1220
+ - vllm/model_executor/layers/fused_moe/
1221
+ - vllm/model_executor/layers/quantization/
1222
- vllm/v1/attention/backends/
1223
+ - vllm/v1/attention/backends/mla/
1224
- vllm/v1/attention/selector.py
1225
+ - .buildkite/scripts/scheduled_integration_test/
1226
- vllm/_aiter_ops.py
1227
- vllm/platforms/rocm.py
1228
optional: true
1229
commands:
1422
- - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-small.txt
1230
+ - bash .buildkite/scripts/scheduled_integration_test/deepseek_v2_lite_prefetch_offload.sh 0.25 200 8030
1231
1232
- label: ":amd: (MI300) MRCR Eval Small Models" # TBD
1233
timeout_in_minutes: 40
@@ -1459,22 +1267,6 @@ steps:
1267
commands:
1268
- pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-small-rocm.txt
1269
1462
-- label: ":amd: (MI300) Multimodal Accuracy Eval (Small Models)" # TBD
1463
- timeout_in_minutes: 35
1464
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1465
- dind: false
1466
- agent_pool: mi300_1
1467
- optional: true
1468
- working_dir: "/vllm-workspace/.buildkite/lm-eval-harness"
1469
- source_file_dependencies:
1470
- - vllm/multimodal/
1471
- - vllm/inputs/
1472
- - vllm/v1/core/
1473
- - vllm/platforms/rocm.py
1474
- - vllm/model_executor/model_loader/
1475
- commands:
1476
- - pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-mm-small.txt --tp-size=1
1477
-
1270
- label: ":amd: (MI300) GPQA Eval (GPT-OSS)" # TBD
1271
timeout_in_minutes: 40
1272
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
@@ -1697,44 +1489,84 @@ steps:
1489
- export VLLM_USE_DEEP_GEMM=0
1490
- pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-large-rocm-tp4.txt --tp-size=4
1491
1700
-#--------------------------------------------------------- mi300 · examples ----------------------------------------------------------#
1492
+#---------------------------------------------------------- mi300 · kernels ----------------------------------------------------------#
1493
1702
-- label: ":amd: (MI300) Examples" # TBD
1703
- timeout_in_minutes: 75
1494
+- label: ":amd: (MI300) Core Operation Kernels Shard %N" # TBD
1495
+ timeout_in_minutes: 40
1496
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1497
dind: false
1498
agent_pool: mi300_1
1707
- optional: true
1708
- working_dir: "/vllm-workspace/examples"
1499
+ parallelism: 3
1500
+ working_dir: "/vllm-workspace/tests"
1501
source_file_dependencies:
1710
- - vllm/entrypoints
1711
- - vllm/multimodal
1712
- - examples/
1502
+ - csrc/
1503
+ - tests/kernels/core
1504
+ - tests/kernels/test_top_k_per_row.py
1505
+ - tests/kernels/test_concat_mla_q.py
1506
+ - tests/kernels/test_rocm_fp32_router_gemm.py
1507
+ - vllm/model_executor/layers/fused_moe/router/gate_linear.py
1508
+ - vllm/model_executor/layers/fused_moe/router/rocm_fp32_router_gemm.py
1509
+ - vllm/model_executor/layers/rotary_embedding/
1510
+ - vllm/_aiter_ops.py
1511
- vllm/platforms/rocm.py
1512
+ - tests/kernels/test_fused_qk_norm_rope_gate.py
1513
+ optional: true
1514
commands:
1715
- # Basic
1716
- # Multi-modal models
1717
- # Pooling models
1718
- # Features demo
1719
- - pip install --no-deps tensorizer
1720
- - python3 basic/offline_inference/chat.py
1721
- - python3 basic/offline_inference/generate.py --model facebook/opt-125m
1722
- - python3 basic/offline_inference/generate.py --model meta-llama/Llama-2-13b-chat-hf --cpu-offload-gb 10
1723
- - python3 basic/offline_inference/classify.py
1724
- - python3 basic/offline_inference/embed.py
1725
- - python3 basic/offline_inference/score.py
1726
- - python3 generate/multimodal/audio_language_offline.py --seed 0
1727
- - python3 generate/multimodal/vision_language_offline.py --seed 0
1728
- - python3 generate/multimodal/vision_language_multi_image_offline.py --seed 0
1729
- - python3 generate/multimodal/encoder_decoder_multimodal_offline.py --model-type whisper --seed 0
1730
- - python3 pooling/embed/vision_embedding_offline.py --seed 0
1731
- - python3 features/automatic_prefix_caching/prefix_caching_offline.py
1732
- - python3 deployment/llm_engine_example.py
1733
- - python3 features/tensorize_vllm_model.py --model facebook/opt-125m serialize --serialized-directory /tmp/ --suffix v1 && python3 features/tensorize_vllm_model.py --model facebook/opt-125m deserialize --path-to-tensors /tmp/vllm/facebook/opt-125m/v1/model.tensors
1734
- - python3 features/speculative_decoding/spec_decode_offline.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
1735
- - 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
1515
+ - pytest -v -s kernels/core --ignore=kernels/core/test_minimax_reduce_rms.py kernels/test_fused_qk_norm_rope_gate.py kernels/test_concat_mla_q.py kernels/test_top_k_per_row.py kernels/test_rocm_fp32_router_gemm.py --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
1516
1737
-#---------------------------------------------------------- mi300 · kernels ----------------------------------------------------------#
1517
+- label: ":amd: (MI300) Quantization Kernels Shard %N"
1518
+ timeout_in_minutes: 120
1519
+ parallelism: 6
1520
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1521
+ dind: false
1522
+ agent_pool: mi300_1
1523
+ optional: true
1524
+ working_dir: "/vllm-workspace/tests"
1525
+ source_file_dependencies:
1526
+ - csrc/quantization/
1527
+ - csrc/rocm/
1528
+ - vllm/model_executor/layers/quantization
1529
+ - tests/kernels/quantization
1530
+ - tests/kernels/quant_utils.py
1531
+ - tests/kernels/utils.py
1532
+ - vllm/_aiter_ops.py
1533
+ - vllm/kernels/aiter_ops.py
1534
+ - vllm/_custom_ops.py
1535
+ - vllm/envs.py
1536
+ - vllm/platforms/rocm.py
1537
+ - vllm/model_executor/kernels/
1538
+ - vllm/v1/attention/backends/rocm_aiter_fa.py
1539
+ - vllm/config/
1540
+ - tests/kernels/quantization/test_rocm_skinny_gemms.py
1541
+ commands:
1542
+ - pytest -v -s kernels/quantization -k "not test_nvfp4_dense_emulation" --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
1543
+
1544
+- label: ":amd: (MI300) MoE Kernels Shard %N" # TBD
1545
+ timeout_in_minutes: 55
1546
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1547
+ dind: false
1548
+ agent_pool: mi300_1
1549
+ optional: true
1550
+ parallelism: 5
1551
+ working_dir: "/vllm-workspace/tests"
1552
+ source_file_dependencies:
1553
+ - csrc/quantization/cutlass_w8a8/moe/
1554
+ - csrc/moe/
1555
+ - csrc/rocm/
1556
+ - tests/kernels/moe
1557
+ - tests/kernels/utils.py
1558
+ - vllm/model_executor/layers/fused_moe/
1559
+ - vllm/model_executor/layers/quantization/mxfp4.py
1560
+ - vllm/model_executor/layers/quantization/utils/quant_utils.py
1561
+ - vllm/distributed/device_communicators/
1562
+ - vllm/envs.py
1563
+ - vllm/config
1564
+ - vllm/_aiter_ops.py
1565
+ - vllm/platforms/rocm.py
1566
+ commands:
1567
+ - pytest -v -s kernels/moe --ignore=kernels/moe/test_modular_oai_triton_moe.py --ignore=kernels/moe/test_modular_kernel_combinations.py --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
1568
+ - pytest -v -s kernels/moe/test_modular_oai_triton_moe.py --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
1569
+ - VLLM_ROCM_USE_AITER=1 VLLM_ROCM_USE_AITER_MOE=1 pytest -v -s kernels/moe/test_modular_kernel_combinations.py --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
1570
1571
- label: ":amd: (MI300) vLLM IR" # TBD
1572
timeout_in_minutes: 50
@@ -1877,60 +1709,6 @@ steps:
1709
commands:
1710
- pytest -v -s kernels/mamba
1711
1880
-- label: ":amd: (MI300) MoE Kernels Shard %N" # TBD
1881
- timeout_in_minutes: 55
1882
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1883
- dind: false
1884
- agent_pool: mi300_1
1885
- optional: true
1886
- parallelism: 5
1887
- working_dir: "/vllm-workspace/tests"
1888
- source_file_dependencies:
1889
- - csrc/quantization/cutlass_w8a8/moe/
1890
- - csrc/moe/
1891
- - csrc/rocm/
1892
- - tests/kernels/moe
1893
- - tests/kernels/utils.py
1894
- - vllm/model_executor/layers/fused_moe/
1895
- - vllm/model_executor/layers/quantization/mxfp4.py
1896
- - vllm/model_executor/layers/quantization/utils/quant_utils.py
1897
- - vllm/distributed/device_communicators/
1898
- - vllm/envs.py
1899
- - vllm/config
1900
- - vllm/_aiter_ops.py
1901
- - vllm/platforms/rocm.py
1902
- commands:
1903
- - pytest -v -s kernels/moe --ignore=kernels/moe/test_modular_oai_triton_moe.py --ignore=kernels/moe/test_modular_kernel_combinations.py --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
1904
- - pytest -v -s kernels/moe/test_modular_oai_triton_moe.py --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
1905
- - VLLM_ROCM_USE_AITER=1 VLLM_ROCM_USE_AITER_MOE=1 pytest -v -s kernels/moe/test_modular_kernel_combinations.py --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
1906
-
1907
-- label: ":amd: (MI300) Quantization Kernels Shard %N"
1908
- timeout_in_minutes: 120
1909
- parallelism: 6
1910
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1911
- dind: false
1912
- agent_pool: mi300_1
1913
- optional: true
1914
- working_dir: "/vllm-workspace/tests"
1915
- source_file_dependencies:
1916
- - csrc/quantization/
1917
- - csrc/rocm/
1918
- - vllm/model_executor/layers/quantization
1919
- - tests/kernels/quantization
1920
- - tests/kernels/quant_utils.py
1921
- - tests/kernels/utils.py
1922
- - vllm/_aiter_ops.py
1923
- - vllm/kernels/aiter_ops.py
1924
- - vllm/_custom_ops.py
1925
- - vllm/envs.py
1926
- - vllm/platforms/rocm.py
1927
- - vllm/model_executor/kernels/
1928
- - vllm/v1/attention/backends/rocm_aiter_fa.py
1929
- - vllm/config/
1930
- - tests/kernels/quantization/test_rocm_skinny_gemms.py
1931
- commands:
1932
- - pytest -v -s kernels/quantization -k "not test_nvfp4_dense_emulation" --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
1933
-
1712
- label: ":amd: (MI300) DeepEP FP8 MoE Kernels" # TBD
1713
timeout_in_minutes: 55
1714
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
@@ -1952,21 +1730,6 @@ steps:
1730
1731
#----------------------------------------------------------- mi300 · lora ------------------------------------------------------------#
1732
1955
-- label: ":amd: (MI300) LoRA Shard %N" # TBD
1956
- timeout_in_minutes: 85
1957
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1958
- dind: false
1959
- agent_pool: mi300_1
1960
- parallelism: 4
1961
- optional: true
1962
- working_dir: "/vllm-workspace/tests"
1963
- source_file_dependencies:
1964
- - vllm/lora
1965
- - tests/lora
1966
- - vllm/platforms/rocm.py
1967
- commands:
1968
- - pytest -v -s lora --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --ignore=lora/test_chatglm3_tp.py --ignore=lora/test_llama_tp.py --ignore=lora/test_qwen3_with_multi_loras.py --ignore=lora/test_olmoe_tp.py --ignore=lora/test_deepseekv2_tp.py --ignore=lora/test_gptoss_tp.py --ignore=lora/test_qwen3moe_tp.py --ignore=lora/test_qwen35_densemodel_lora.py
1969
-
1733
- label: ":amd: (MI300) LoRA TP (Distributed) %N"
1734
timeout_in_minutes: 65
1735
parallelism: 4
@@ -2035,26 +1798,6 @@ steps:
1798
commands:
1799
- pytest -v -s models/test_initialization.py::test_can_initialize_small_subset
1800
2038
-- label: ":amd: (MI300) Basic Models (Other)" # TBD
2039
- timeout_in_minutes: 65
2040
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
2041
- dind: false
2042
- agent_pool: mi300_1
2043
- optional: true
2044
- working_dir: "/vllm-workspace/tests"
2045
- source_file_dependencies:
2046
- - vllm/
2047
- - tests/models/test_terratorch.py
2048
- - tests/models/transformers/test_backend.py
2049
- - tests/models/test_registry.py
2050
- - tests/models/test_hyv4_rocm.py
2051
- - tests/models/test_vision.py
2052
- - tests/models/test_deepseek_v4_vl_rocm.py
2053
- commands:
2054
- - pytest -v -s models/test_terratorch.py models/transformers/test_backend.py models/test_registry.py models/test_deepseek_v4_vl_rocm.py
2055
- - VLLM_ROCM_USE_AITER=1 pytest -v -s models/test_hyv4_rocm.py -m 'not distributed'
2056
- - pytest -v -s models/test_vision.py::test_simple_mrope_vision_model_spatial_merge
2057
-
1801
#----------------------------------------------------- mi300 · models / language -----------------------------------------------------#
1802
1803
- label: ":amd: (MI300) Language Models (MTEB)" # TBD
@@ -2084,43 +1827,6 @@ steps:
1827
commands:
1828
- pytest -v -s models/language/pooling -m 'not core_model' --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
1829
2087
-- label: ":amd: (MI300) Language Models (Standard)" # TBD
2088
- timeout_in_minutes: 60
2089
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
2090
- dind: false
2091
- agent_pool: mi300_1
2092
- optional: true
2093
- working_dir: "/vllm-workspace/tests"
2094
- source_file_dependencies:
2095
- - vllm/
2096
- - tests/models/language
2097
- commands:
2098
- - pip freeze | grep -E 'torch'
2099
- - pytest -v -s models/language -m 'core_model and (not slow_test)'
2100
-
2101
-- label: ":amd: (MI300) Language Models (Extra Standard) Shard %N" # TBD
2102
- timeout_in_minutes: 40
2103
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
2104
- dind: false
2105
- agent_pool: mi300_1
2106
- optional: true
2107
- parallelism: 2
2108
- working_dir: "/vllm-workspace/tests"
2109
- source_file_dependencies:
2110
- - vllm/model_executor/models/
2111
- - vllm/model_executor/model_loader/
2112
- - vllm/model_executor/layers/
2113
- - vllm/v1/attention/backends/
2114
- - vllm/v1/attention/selector.py
2115
- - tests/models/language/pooling/test_embedding.py
2116
- - tests/models/language/generation/test_common.py
2117
- - tests/models/language/pooling/test_classification.py
2118
- - vllm/_aiter_ops.py
2119
- - vllm/platforms/rocm.py
2120
- commands:
2121
- - pip freeze | grep -E 'torch'
2122
- - pytest -v -s models/language -m 'core_model and slow_test' --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
2123
-
1830
- label: ":amd: (MI300) Language Models (Extended Generation)" # TBD
1831
timeout_in_minutes: 95
1832
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
@@ -2181,20 +1887,6 @@ steps:
1887
commands:
1888
- pytest -v -s models/multimodal/generation/test_common.py -m 'split(group=1) and not core_model'
1889
2184
-- label: ":amd: (MI300) Multimodal Models (Standard) 1: qwen2" # TBD
2185
- timeout_in_minutes: 65
2186
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
2187
- dind: false
2188
- agent_pool: mi300_1
2189
- optional: true
2190
- working_dir: "/vllm-workspace/tests"
2191
- source_file_dependencies:
2192
- - vllm/
2193
- - tests/models/multimodal
2194
- commands:
2195
- - pytest -v -s models/multimodal/generation/test_common.py -m core_model -k "qwen2"
2196
- - pytest -v -s models/multimodal/generation/test_ultravox.py -m core_model
2197
-
1890
- label: ":amd: (MI300) Multimodal Models (Standard) 3: llava + qwen2_vl" # TBD
1891
timeout_in_minutes: 55
1892
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
@@ -2239,21 +1931,6 @@ steps:
1931
commands:
1932
- pytest -v -s models/multimodal/pooling -m 'not core_model'
1933
2242
-- label: ":amd: (MI300) Multimodal Models (Standard) 2: qwen3 + gemma" # TBD
2243
- timeout_in_minutes: 55
2244
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
2245
- dind: false
2246
- agent_pool: mi300_1
2247
- optional: true
2248
- working_dir: "/vllm-workspace/tests"
2249
- source_file_dependencies:
2250
- - vllm/
2251
- - tests/models/multimodal
2252
- commands:
2253
- - pytest -v -s models/multimodal/generation/test_common.py -m core_model -k "qwen3 or gemma"
2254
- - pytest -v -s models/multimodal/generation/test_mm_prefix_lm.py -m core_model
2255
- - pytest -v -s models/multimodal/generation/test_qwen2_5_vl.py -m core_model
2256
-
1934
#----------------------------------------------------- mi300 · models / quantized -----------------------------------------------------#
1935
1936
- label: ":amd: (MI300) Quantized Models" # TBD
@@ -2650,8 +2327,8 @@ steps:
2327
2328
#------------------------------------------------------------ mi300 · v1 -------------------------------------------------------------#
2329
2653
-- label: ":amd: (MI300) Acceptance Length (Large Models)" # TBD
2654
- timeout_in_minutes: 45
2330
+- label: ":amd: (MI300) Spec Decode Speculators + MTP" # TBD
2331
+ timeout_in_minutes: 75
2332
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
2333
dind: false
2334
agent_pool: mi300_1
@@ -2659,28 +2336,43 @@ steps:
2336
working_dir: "/vllm-workspace/tests"
2337
source_file_dependencies:
2338
- vllm/v1/spec_decode/
2662
- - vllm/model_executor/models/mlp_speculator.py
2663
- - tests/v1/spec_decode/test_acceptance_length.py
2339
+ - vllm/v1/worker/gpu/spec_decode/
2340
+ - vllm/model_executor/model_loader/
2341
+ - vllm/v1/sample/
2342
+ - vllm/model_executor/layers/
2343
+ - vllm/transformers_utils/configs/speculators/
2344
+ - tests/v1/e2e/spec_decode/
2345
- vllm/platforms/rocm.py
2346
+ - vllm/v1/attention/backends/
2347
commands:
2666
- - export VLLM_ALLOW_INSECURE_SERIALIZATION=1
2667
- - pytest -v -s v1/spec_decode/test_acceptance_length.py -m slow_test
2348
+ - pytest -v -s v1/e2e/spec_decode/speculators/
2349
+ - pytest -v -s v1/e2e/spec_decode/mtp/
2350
2669
-- label: ":amd: (MI300) E2E Scheduling" # TBD
2670
- timeout_in_minutes: 55
2351
+- label: ":amd: (MI300) V1 Spec Decode" # TBD
2352
+ timeout_in_minutes: 50
2353
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
2354
dind: false
2355
agent_pool: mi300_1
2674
- optional: true
2356
working_dir: "/vllm-workspace/tests"
2357
source_file_dependencies:
2358
+ - vllm/
2359
+ - tests/v1/spec_decode
2360
+ - vllm/config/
2361
+ - vllm/distributed/
2362
+ - vllm/inputs/
2363
+ - vllm/model_executor/
2364
+ - vllm/models/qwen4_exp/
2365
+ - vllm/platforms/
2366
+ - vllm/sampling_params.py
2367
+ - vllm/transformers_utils/
2368
+ - vllm/utils/
2369
- vllm/v1/
2678
- - tests/v1/e2e/general/
2679
- - vllm/platforms/rocm.py
2370
+ optional: true
2371
commands:
2681
- - pytest -v -s v1/e2e/general/test_async_scheduling.py
2372
+ - export VLLM_WORKER_MULTIPROC_METHOD=spawn
2373
+ - pytest -v -s -m 'not slow_test' v1/spec_decode
2374
2683
-- label: ":amd: (MI300) V1 Engine" # TBD
2375
+- label: ":amd: (MI300) Acceptance Length (Large Models)" # TBD
2376
timeout_in_minutes: 45
2377
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
2378
dind: false
@@ -2688,110 +2380,44 @@ steps:
2380
optional: true
2381
working_dir: "/vllm-workspace/tests"
2382
source_file_dependencies:
2691
- - vllm/v1/engine/
2692
- - tests/v1/engine/
2693
- - vllm/platforms/rocm.py
2694
- - tests/v1/test_tensor_ipc_queue.py
2695
- commands:
2696
- - pytest -v -s v1/engine/test_preprocess_error_handling.py
2697
- - pytest -v -s v1/engine --ignore v1/engine/test_preprocess_error_handling.py
2698
- - pytest -v -s v1/test_tensor_ipc_queue.py
2699
-
2700
-- label: ":amd: (MI300) Spec Decode Draft Model" # TBD
2701
- timeout_in_minutes: 60
2702
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
2703
- dind: false
2704
- agent_pool: mi300_1
2705
- optional: true
2706
- working_dir: "/vllm-workspace/tests"
2707
- source_file_dependencies:
2708
- - vllm/v1/spec_decode/
2709
- - vllm/v1/worker/gpu/spec_decode/
2710
- - vllm/model_executor/model_loader/
2711
- - vllm/v1/sample/
2712
- - vllm/model_executor/layers/
2713
- - tests/v1/e2e/spec_decode/
2714
- - vllm/platforms/rocm.py
2715
- commands:
2716
- - pytest -v -s v1/e2e/spec_decode/draft_model/
2717
-
2718
-- label: ":amd: (MI300) Spec Decode Eagle 1: DeepSeek + Qwen"
2719
- timeout_in_minutes: 60
2720
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
2721
- dind: false
2722
- agent_pool: mi300_1
2723
- optional: true
2724
- working_dir: "/vllm-workspace/tests"
2725
- source_file_dependencies:
2726
- - vllm/v1/spec_decode/
2727
- - vllm/v1/worker/gpu/spec_decode/
2728
- - vllm/model_executor/model_loader/
2729
- - vllm/v1/sample/
2730
- - vllm/model_executor/layers/
2731
- - tests/v1/e2e/spec_decode/
2732
- - vllm/platforms/rocm.py
2733
- commands:
2734
- - pytest -v -s v1/e2e/spec_decode/eagle/ -k "deepseek_eagle or qwen3_eagle3"
2735
-
2736
-- label: ":amd: (MI300) Spec Decode Eagle 2: Llama 3 + Qwen VL + Other"
2737
- timeout_in_minutes: 60
2738
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
2739
- dind: false
2740
- agent_pool: mi300_1
2741
- optional: true
2742
- working_dir: /vllm-workspace/tests
2743
- source_file_dependencies:
2383
- vllm/v1/spec_decode/
2745
- - vllm/v1/worker/gpu/spec_decode/
2746
- - vllm/model_executor/model_loader/
2747
- - vllm/v1/sample/
2748
- - vllm/model_executor/layers/
2749
- - tests/v1/e2e/spec_decode/
2384
+ - vllm/model_executor/models/mlp_speculator.py
2385
+ - tests/v1/spec_decode/test_acceptance_length.py
2386
- vllm/platforms/rocm.py
2387
commands:
2752
- - pytest -v -s v1/e2e/spec_decode/eagle/ -k "not deepseek_eagle and not qwen3_eagle3"
2753
-
2388
+ - export VLLM_ALLOW_INSECURE_SERIALIZATION=1
2389
+ - pytest -v -s v1/spec_decode/test_acceptance_length.py -m slow_test
2390
2755
-- label: ":amd: (MI300) Spec Decode N-Gram + Suffix" # TBD
2756
- timeout_in_minutes: 35
2391
+- label: ":amd: (MI300) E2E Scheduling" # TBD
2392
+ timeout_in_minutes: 55
2393
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
2394
dind: false
2395
agent_pool: mi300_1
2396
optional: true
2397
working_dir: "/vllm-workspace/tests"
2398
source_file_dependencies:
2763
- - vllm/v1/spec_decode/
2764
- - vllm/v1/worker/gpu/spec_decode/
2765
- - vllm/model_executor/model_loader/
2766
- - vllm/v1/sample/
2767
- - vllm/model_executor/layers/
2768
- - tests/v1/e2e/spec_decode/
2769
- - tests/spec_decode/
2399
+ - vllm/v1/
2400
+ - tests/v1/e2e/general/
2401
- vllm/platforms/rocm.py
2402
commands:
2772
- - pytest -v -s v1/e2e/spec_decode/ngram_suffix/
2773
- - python3 spec_decode/test_custom_proposer.py
2403
+ - pytest -v -s v1/e2e/general/test_async_scheduling.py
2404
2775
-- label: ":amd: (MI300) Spec Decode Speculators + MTP" # TBD
2776
- timeout_in_minutes: 75
2405
+- label: ":amd: (MI300) V1 Engine" # TBD
2406
+ timeout_in_minutes: 45
2407
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
2408
dind: false
2409
agent_pool: mi300_1
2410
optional: true
2411
working_dir: "/vllm-workspace/tests"
2412
source_file_dependencies:
2783
- - vllm/v1/spec_decode/
2784
- - vllm/v1/worker/gpu/spec_decode/
2785
- - vllm/model_executor/model_loader/
2786
- - vllm/v1/sample/
2787
- - vllm/model_executor/layers/
2788
- - vllm/transformers_utils/configs/speculators/
2789
- - tests/v1/e2e/spec_decode/
2413
+ - vllm/v1/engine/
2414
+ - tests/v1/engine/
2415
- vllm/platforms/rocm.py
2791
- - vllm/v1/attention/backends/
2416
+ - tests/v1/test_tensor_ipc_queue.py
2417
commands:
2793
- - pytest -v -s v1/e2e/spec_decode/speculators/
2794
- - pytest -v -s v1/e2e/spec_decode/mtp/
2418
+ - pytest -v -s v1/engine/test_preprocess_error_handling.py
2419
+ - pytest -v -s v1/engine --ignore v1/engine/test_preprocess_error_handling.py
2420
+ - pytest -v -s v1/test_tensor_ipc_queue.py
2421
2422
- label: ":amd: (MI300) Speculators Correctness Nightly"
2423
timeout_in_minutes: 40
@@ -2826,30 +2452,6 @@ steps:
2452
- export VLLM_ALLOW_INSECURE_SERIALIZATION=1
2453
- pytest -v -s v1/spec_decode/test_speculators_correctness.py -m slow_test
2454
2829
-- label: ":amd: (MI300) V1 Spec Decode" # TBD
2830
- timeout_in_minutes: 50
2831
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
2832
- dind: false
2833
- agent_pool: mi300_1
2834
- working_dir: "/vllm-workspace/tests"
2835
- source_file_dependencies:
2836
- - vllm/
2837
- - tests/v1/spec_decode
2838
- - vllm/config/
2839
- - vllm/distributed/
2840
- - vllm/inputs/
2841
- - vllm/model_executor/
2842
- - vllm/models/qwen4_exp/
2843
- - vllm/platforms/
2844
- - vllm/sampling_params.py
2845
- - vllm/transformers_utils/
2846
- - vllm/utils/
2847
- - vllm/v1/
2848
- optional: true
2849
- commands:
2850
- - export VLLM_WORKER_MULTIPROC_METHOD=spawn
2851
- - pytest -v -s -m 'not slow_test' v1/spec_decode
2852
-
2455
- label: ":amd: (MI300) Extract Hidden States Integration" # TBD
2456
timeout_in_minutes: 30
2457
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
@@ -2904,359 +2506,82 @@ steps:
2506
commands:
2507
- pytest -v -s v1/attention --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
2508
2907
-- label: ":amd: (MI300) V1 Core"
2908
- timeout_in_minutes: 60
2509
+- label: ":amd: (MI300) Metrics, Tracing" # TBD
2510
+ timeout_in_minutes: 30
2511
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
2512
dind: false
2911
- agent_pool: mi300_1
2513
+ agent_pool: mi300_2
2514
+ optional: true
2515
+ num_gpus: 2
2516
working_dir: "/vllm-workspace/tests"
2517
source_file_dependencies:
2518
- vllm/
2915
- - tests/v1/core
2916
- - tests/v1/executor
2917
- - tests/v1/kv_offload
2918
- - tests/v1/simple_kv_offload
2919
- - tests/v1/worker
2920
- - tests/v1/streaming_input
2921
- - tests/v1/kv_connector/unit
2922
- - tests/v1/ec_connector/unit
2923
- - tests/v1/metrics
2924
- - tests/entrypoints/openai/correctness/test_lmeval.py
2519
+ - tests/v1/tracing
2520
+ - tests/tracing/
2521
- vllm/config/
2522
- vllm/distributed/
2523
- vllm/engine/
2928
- - vllm/entrypoints/pooling/
2524
- vllm/inputs/
2930
- - vllm/lora/
2525
- vllm/model_executor/
2526
- vllm/multimodal/
2933
- - vllm/outputs.py
2527
- vllm/platforms/
2935
- - vllm/pooling_params.py
2936
- - vllm/profiler/
2528
- vllm/sampling_params.py
2938
- - vllm/tokenizers/
2529
+ - vllm/tracing/
2530
- vllm/transformers_utils/
2531
- vllm/utils/
2532
- vllm/v1/
2942
- optional: true
2533
commands:
2944
- # - export HSA_NO_SCRATCH_RECLAIM=1
2945
- - export VLLM_WORKER_MULTIPROC_METHOD=spawn
2946
- - pytest -v -s -m 'not cpu_test' v1/core
2534
+ - "pip install \
2535
+ 'opentelemetry-sdk>=1.26.0' \
2536
+ 'opentelemetry-api>=1.26.0' \
2537
+ 'opentelemetry-exporter-otlp>=1.26.0' \
2538
+ 'opentelemetry-semantic-conventions-ai>=0.4.1'"
2539
+ - pytest -v -s v1/tracing
2540
+ - pytest -v -s tracing
2541
2948
-- label: ":amd: (MI300) V1 Executor + Worker"
2949
- timeout_in_minutes: 60
2542
+- label: ":amd: (MI300) V1 E2E" # TBD
2543
+ timeout_in_minutes: 30
2544
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
2545
dind: false
2952
- agent_pool: mi300_1
2953
- working_dir: /vllm-workspace/tests
2954
- source_file_dependencies:
2955
- - vllm/
2956
- - tests/v1/core
2957
- - tests/v1/executor
2958
- - tests/v1/kv_offload
2959
- - tests/v1/simple_kv_offload
2960
- - tests/v1/worker
2961
- - tests/v1/streaming_input
2962
- - tests/v1/kv_connector/unit
2963
- - tests/v1/ec_connector/unit
2964
- - tests/v1/metrics
2965
- - tests/entrypoints/openai/correctness/test_lmeval.py
2966
- - vllm/config/
2967
- - vllm/distributed/
2968
- - vllm/engine/
2969
- - vllm/entrypoints/pooling/
2970
- - vllm/inputs/
2971
- - vllm/lora/
2972
- - vllm/model_executor/
2973
- - vllm/multimodal/
2974
- - vllm/outputs.py
2975
- - vllm/platforms/
2976
- - vllm/pooling_params.py
2977
- - vllm/profiler/
2978
- - vllm/sampling_params.py
2979
- - vllm/tokenizers/
2980
- - vllm/transformers_utils/
2981
- - vllm/utils/
2982
- - vllm/v1/
2983
- commands:
2984
- - export VLLM_WORKER_MULTIPROC_METHOD=spawn
2985
- - pytest -v -s v1/executor
2986
- - pytest -v -s v1/worker
2987
- - pytest -v -s v1/streaming_input
2546
+ agent_pool: mi300_2
2547
+ num_gpus: 2
2548
optional: true
2989
-
2990
-- label: ":amd: (MI300) V1 KV Offload"
2991
- timeout_in_minutes: 60
2992
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
2993
- dind: false
2994
- agent_pool: mi300_1
2995
- working_dir: /vllm-workspace/tests
2549
+ working_dir: "/vllm-workspace/tests"
2550
source_file_dependencies:
2551
- vllm/
2998
- - tests/v1/core
2999
- - tests/v1/executor
3000
- - tests/v1/kv_offload
3001
- - tests/v1/simple_kv_offload
3002
- - tests/v1/worker
3003
- - tests/v1/streaming_input
3004
- - tests/v1/kv_connector/unit
3005
- - tests/v1/ec_connector/unit
3006
- - tests/v1/metrics
3007
- - tests/entrypoints/openai/correctness/test_lmeval.py
2552
+ - tests/v1/e2e
2553
+ - vllm/compilation/
2554
- vllm/config/
2555
- vllm/distributed/
2556
- vllm/engine/
3011
- - vllm/entrypoints/pooling/
2557
+ - vllm/envs.py
2558
+ - vllm/forward_context.py
2559
- vllm/inputs/
3013
- - vllm/lora/
2560
+ - vllm/logger.py
2561
+ - vllm/logging_utils/
2562
- vllm/model_executor/
2563
- vllm/multimodal/
3016
- - vllm/outputs.py
2564
- vllm/platforms/
3018
- - vllm/pooling_params.py
3019
- - vllm/profiler/
2565
- vllm/sampling_params.py
3021
- - vllm/tokenizers/
2566
- vllm/transformers_utils/
2567
+ - vllm/triton_utils/
2568
- vllm/utils/
2569
- vllm/v1/
2570
+ - tests/v1/e2e/spec_decode/
2571
commands:
3026
- - export VLLM_WORKER_MULTIPROC_METHOD=spawn
3027
- - bash /vllm-workspace/.buildkite/scripts/install-kv-offload.sh
3028
- - pytest -v -s v1/kv_offload
3029
- - pytest -v -s v1/simple_kv_offload
3030
- optional: true
2572
+ - >-
2573
+ pytest -v -s
2574
+ v1/e2e/spec_decode/draft_model/test_draft_model.py::test_draft_model_tensor_parallelism
2575
+ v1/e2e/spec_decode/draft_model/test_draft_model.py::test_draft_model_engine_args_tensor_parallelism
2576
3032
-- label: ":amd: (MI300) V1 KV Connectors Shard %N"
2577
+- label: ":amd: (MI300) Mooncake EC TCP E2E"
2578
timeout_in_minutes: 60
3034
- parallelism: 4
2579
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
2580
dind: false
3037
- agent_pool: mi300_1
3038
- working_dir: /vllm-workspace/tests
3039
- source_file_dependencies:
3040
- - vllm/
3041
- - tests/v1/core
3042
- - tests/v1/executor
3043
- - tests/v1/kv_offload
3044
- - tests/v1/simple_kv_offload
3045
- - tests/v1/worker
3046
- - tests/v1/streaming_input
3047
- - tests/v1/kv_connector/unit
3048
- - tests/v1/ec_connector/unit
3049
- - tests/v1/metrics
3050
- - tests/entrypoints/openai/correctness/test_lmeval.py
3051
- - vllm/config/
3052
- - vllm/distributed/
3053
- - vllm/engine/
3054
- - vllm/entrypoints/pooling/
3055
- - vllm/inputs/
3056
- - vllm/lora/
3057
- - vllm/model_executor/
3058
- - vllm/multimodal/
3059
- - vllm/outputs.py
3060
- - vllm/platforms/
3061
- - vllm/pooling_params.py
3062
- - vllm/profiler/
3063
- - vllm/sampling_params.py
3064
- - vllm/tokenizers/
3065
- - vllm/transformers_utils/
3066
- - vllm/utils/
3067
- - vllm/v1/
3068
- commands:
3069
- - bash /vllm-workspace/.buildkite/scripts/install-kv-connectors.sh
3070
- - export VLLM_WORKER_MULTIPROC_METHOD=spawn
3071
- - pytest -v -s -m 'not cpu_test' v1/kv_connector/unit --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
3072
- - pytest -v -s -m 'not cpu_test' v1/ec_connector/unit --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
3073
- optional: true
3074
-
3075
-- label: ":amd: (MI300) V1 Metrics + LM Eval"
3076
- timeout_in_minutes: 65
3077
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
3078
- dind: false
3079
- agent_pool: mi300_1
3080
- working_dir: /vllm-workspace/tests
3081
- source_file_dependencies:
3082
- - vllm/
3083
- - tests/v1/core
3084
- - tests/v1/executor
3085
- - tests/v1/kv_offload
3086
- - tests/v1/simple_kv_offload
3087
- - tests/v1/worker
3088
- - tests/v1/streaming_input
3089
- - tests/v1/kv_connector/unit
3090
- - tests/v1/ec_connector/unit
3091
- - tests/v1/metrics
3092
- - tests/entrypoints/openai/correctness/test_lmeval.py
3093
- - vllm/config/
3094
- - vllm/distributed/
3095
- - vllm/engine/
3096
- - vllm/entrypoints/pooling/
3097
- - vllm/inputs/
3098
- - vllm/lora/
3099
- - vllm/model_executor/
3100
- - vllm/multimodal/
3101
- - vllm/outputs.py
3102
- - vllm/platforms/
3103
- - vllm/pooling_params.py
3104
- - vllm/profiler/
3105
- - vllm/sampling_params.py
3106
- - vllm/tokenizers/
3107
- - vllm/transformers_utils/
3108
- - vllm/utils/
3109
- - vllm/v1/
3110
- commands:
3111
- - export GIT_CONFIG_COUNT=1
3112
- - export GIT_CONFIG_KEY_0=http.version
3113
- - export GIT_CONFIG_VALUE_0=HTTP/1.1
3114
- - export GIT_TERMINAL_PROMPT=0
3115
- - export VLLM_WORKER_MULTIPROC_METHOD=spawn
3116
- - pytest -v -s -m 'not cpu_test' v1/metrics
3117
- - pip install -U git+https://github.com/vllm-project/lm-evaluation-harness.git@streaming-api
3118
- - pytest -v -s entrypoints/openai/correctness/test_lmeval.py::test_lm_eval_accuracy_v1_engine
3119
- optional: true
3120
-
3121
-
3122
-- label: ":amd: (MI300) V1 Sample"
3123
- timeout_in_minutes: 70
3124
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
3125
- dind: false
3126
- agent_pool: mi300_1
3127
- optional: true
3128
- working_dir: "/vllm-workspace/tests"
3129
- source_file_dependencies:
3130
- - vllm/
3131
- - tests/v1/sample
3132
- - tests/v1/logits_processors
3133
- - tests/v1/test_oracle.py
3134
- - tests/v1/test_request.py
3135
- - tests/v1/test_outputs.py
3136
- - vllm/config/
3137
- - vllm/distributed/
3138
- - vllm/engine/
3139
- - vllm/inputs/
3140
- - vllm/logger.py
3141
- - vllm/model_executor/
3142
- - vllm/platforms/
3143
- - vllm/sampling_params.py
3144
- - vllm/transformers_utils/
3145
- - vllm/utils/
3146
- - vllm/v1/
3147
- commands:
3148
- - export VLLM_WORKER_MULTIPROC_METHOD=spawn
3149
- - pytest -v -s v1/sample
3150
-
3151
-- label: ":amd: (MI300) V1 Logits + Oracle"
3152
- timeout_in_minutes: 70
3153
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
3154
- dind: false
3155
- agent_pool: mi300_1
3156
- optional: true
3157
- working_dir: /vllm-workspace/tests
3158
- source_file_dependencies:
3159
- - vllm/
3160
- - tests/v1/sample
3161
- - tests/v1/logits_processors
3162
- - tests/v1/test_oracle.py
3163
- - tests/v1/test_request.py
3164
- - tests/v1/test_outputs.py
3165
- - vllm/config/
3166
- - vllm/distributed/
3167
- - vllm/engine/
3168
- - vllm/inputs/
3169
- - vllm/logger.py
3170
- - vllm/model_executor/
3171
- - vllm/platforms/
3172
- - vllm/sampling_params.py
3173
- - vllm/transformers_utils/
3174
- - vllm/utils/
3175
- - vllm/v1/
3176
- commands:
3177
- - export VLLM_WORKER_MULTIPROC_METHOD=spawn
3178
- - pytest -v -s v1/logits_processors
3179
- - pytest -v -s v1/test_oracle.py
3180
- - pytest -v -s v1/test_request.py
3181
- - pytest -v -s v1/test_outputs.py
3182
-
3183
-
3184
-- label: ":amd: (MI300) Metrics, Tracing" # TBD
3185
- timeout_in_minutes: 30
3186
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
3187
- dind: false
3188
- agent_pool: mi300_2
3189
- optional: true
3190
- num_gpus: 2
3191
- working_dir: "/vllm-workspace/tests"
3192
- source_file_dependencies:
3193
- - vllm/
3194
- - tests/v1/tracing
3195
- - tests/tracing/
3196
- - vllm/config/
3197
- - vllm/distributed/
3198
- - vllm/engine/
3199
- - vllm/inputs/
3200
- - vllm/model_executor/
3201
- - vllm/multimodal/
3202
- - vllm/platforms/
3203
- - vllm/sampling_params.py
3204
- - vllm/tracing/
3205
- - vllm/transformers_utils/
3206
- - vllm/utils/
3207
- - vllm/v1/
3208
- commands:
3209
- - "pip install \
3210
- 'opentelemetry-sdk>=1.26.0' \
3211
- 'opentelemetry-api>=1.26.0' \
3212
- 'opentelemetry-exporter-otlp>=1.26.0' \
3213
- 'opentelemetry-semantic-conventions-ai>=0.4.1'"
3214
- - pytest -v -s v1/tracing
3215
- - pytest -v -s tracing
3216
-
3217
-- label: ":amd: (MI300) V1 E2E" # TBD
3218
- timeout_in_minutes: 30
3219
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
3220
- dind: false
3221
- agent_pool: mi300_2
3222
- num_gpus: 2
3223
- optional: true
3224
- working_dir: "/vllm-workspace/tests"
3225
- source_file_dependencies:
3226
- - vllm/
3227
- - tests/v1/e2e
3228
- - vllm/compilation/
3229
- - vllm/config/
3230
- - vllm/distributed/
3231
- - vllm/engine/
3232
- - vllm/envs.py
3233
- - vllm/forward_context.py
3234
- - vllm/inputs/
3235
- - vllm/logger.py
3236
- - vllm/logging_utils/
3237
- - vllm/model_executor/
3238
- - vllm/multimodal/
3239
- - vllm/platforms/
3240
- - vllm/sampling_params.py
3241
- - vllm/transformers_utils/
3242
- - vllm/triton_utils/
3243
- - vllm/utils/
3244
- - vllm/v1/
3245
- - tests/v1/e2e/spec_decode/
3246
- commands:
3247
- - >-
3248
- pytest -v -s
3249
- v1/e2e/spec_decode/draft_model/test_draft_model.py::test_draft_model_tensor_parallelism
3250
- v1/e2e/spec_decode/draft_model/test_draft_model.py::test_draft_model_engine_args_tensor_parallelism
3251
-
3252
-- label: ":amd: (MI300) Mooncake EC TCP E2E"
3253
- timeout_in_minutes: 60
3254
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
3255
- dind: false
3256
- agent_pool: mi300_2
3257
- num_gpus: 2
3258
- optional: true
3259
- working_dir: "/vllm-workspace"
2581
+ agent_pool: mi300_2
2582
+ num_gpus: 2
2583
+ optional: true
2584
+ working_dir: "/vllm-workspace"
2585
source_file_dependencies:
2586
- vllm/distributed/ec_transfer/
2587
- vllm/config/ec_transfer.py
@@ -3342,7 +2667,7 @@ steps:
2667
- pytest -v -s distributed/test_utils.py
2668
2669
- label: ":amd: (MI300) Hybrid SSM NixlConnector PD accuracy" # TBD
3345
- timeout_in_minutes: 60
2670
+ timeout_in_minutes: 80
2671
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
2672
dind: false
2673
agent_pool: mi300_4
@@ -3435,8 +2760,43 @@ steps:
2760
# #
2761
#########################################################################################################################################
2762
2763
+#----------------------------------------------------- mi355 · basic_correctness -----------------------------------------------------#
2764
+
2765
+- label: ":amd: (MI355 DPX) Basic Correctness" # TBD
2766
+ timeout_in_minutes: 60
2767
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
2768
+ dind: false
2769
+ agent_pool: mi355_dpx
2770
+ fast_check: true
2771
+ working_dir: "/vllm-workspace/tests"
2772
+ source_file_dependencies:
2773
+ - vllm/
2774
+ - tests/basic_correctness/test_basic_correctness
2775
+ - tests/basic_correctness/test_cpu_offload
2776
+ - tests/basic_correctness/test_mem.py
2777
+ optional: true
2778
+ commands:
2779
+ - export VLLM_WORKER_MULTIPROC_METHOD=spawn
2780
+ - pytest -v -s basic_correctness/test_mem.py
2781
+ - VLLM_TARGET_TEST_SUITE=MI355 pytest -v -s basic_correctness/test_basic_correctness.py
2782
+ - pytest -v -s basic_correctness/test_cpu_offload.py
2783
+
2784
#-------------------------------------------------------- mi355 · benchmarks ---------------------------------------------------------#
2785
2786
+- label: ":amd: (MI355 DPX) Benchmarks CLI" # TBD
2787
+ timeout_in_minutes: 40
2788
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
2789
+ dind: false
2790
+ agent_pool: mi355_dpx
2791
+ optional: true
2792
+ working_dir: "/vllm-workspace/tests"
2793
+ source_file_dependencies:
2794
+ - vllm/
2795
+ - tests/benchmarks/
2796
+ - rust/src/bench/tests/python_serve_flags.txt
2797
+ commands:
2798
+ - pytest -v -s benchmarks/
2799
+
2800
- label: ":amd: (MI355) Attention Benchmark Smoke" # TBD
2801
timeout_in_minutes: 25
2802
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
@@ -3454,11 +2814,11 @@ steps:
2814
2815
#---------------------------------------------------------- mi355 · compile ----------------------------------------------------------#
2816
3457
-- label: ":amd: (MI355) PyTorch Compilation H100 Cases"
2817
+- label: ":amd: (MI355 DPX) PyTorch Compilation H100 Cases"
2818
timeout_in_minutes: 45
2819
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
2820
dind: false
3461
- agent_pool: mi355_1
2821
+ agent_pool: mi355_dpx
2822
num_gpus: 1
2823
optional: true
2824
working_dir: "/vllm-workspace/tests"
@@ -3519,11 +2879,27 @@ steps:
2879
2880
#-------------------------------------------------------- mi355 · distributed --------------------------------------------------------#
2881
3522
-- label: ":amd: (MI355) Sharded RDT Weight Transfer"
2882
+- label: ":amd: (MI355 DPX) EPLB Algorithm" # TBD
2883
+ timeout_in_minutes: 30
2884
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
2885
+ dind: false
2886
+ agent_pool: mi355_dpx
2887
+ optional: true
2888
+ working_dir: "/vllm-workspace/tests"
2889
+ source_file_dependencies:
2890
+ - vllm/distributed/eplb
2891
+ - tests/distributed/test_eplb_algo.py
2892
+ - tests/distributed/test_eplb_utils.py
2893
+ - vllm/platforms/rocm.py
2894
+ commands:
2895
+ - pytest -v -s distributed/test_eplb_algo.py
2896
+ - pytest -v -s distributed/test_eplb_utils.py
2897
+
2898
+- label: ":amd: (MI355 DPX) Sharded RDT Weight Transfer"
2899
timeout_in_minutes: 45
2900
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
2901
dind: false
3526
- agent_pool: mi355_1
2902
+ agent_pool: mi355_dpx
2903
num_gpus: 1
2904
optional: true
2905
working_dir: "/vllm-workspace/tests"
@@ -3852,14 +3228,63 @@ steps:
3228
DECODER_KV_LAYOUT=
3229
bash v1/kv_connector/nixl_integration/run_accuracy_test.sh
3230
3231
+#---------------------------------------------------------- mi355 · engine -----------------------------------------------------------#
3232
+
3233
+- label: ":amd: (MI355 DPX) Engine Core" # TBD
3234
+ timeout_in_minutes: 40
3235
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
3236
+ dind: false
3237
+ agent_pool: mi355_dpx
3238
+ optional: true
3239
+ working_dir: "/vllm-workspace/tests"
3240
+ source_file_dependencies:
3241
+ - vllm/
3242
+ - tests/engine
3243
+ - tests/test_sequence
3244
+ - tests/test_config
3245
+ - tests/test_logger
3246
+ - tests/test_vllm_port
3247
+ - tests/jit_monitor/test_hooks.py
3248
+ - tests/jit_monitor/test_hooks_gpu.py
3249
+ - vllm/compilation/
3250
+ - vllm/config/
3251
+ - vllm/engine/
3252
+ - vllm/entrypoints/logger.py
3253
+ - vllm/envs.py
3254
+ - vllm/logger.py
3255
+ - vllm/logging_utils/
3256
+ - vllm/platforms/
3257
+ - vllm/sequence.py
3258
+ - vllm/triton_utils/
3259
+ - vllm/utils/
3260
+ commands:
3261
+ - 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
3262
+
3263
#-------------------------------------------------------- mi355 · entrypoints --------------------------------------------------------#
3264
3857
-- label: ":amd: (MI355) Entrypoints Integration (API Server) %N"
3265
+- label: ":amd: (MI355 DPX) Entrypoints Integration (LLM)" # TBD
3266
+ timeout_in_minutes: 90
3267
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
3268
+ dind: false
3269
+ agent_pool: mi355_dpx
3270
+ optional: true
3271
+ fast_check: true
3272
+ working_dir: "/vllm-workspace/tests"
3273
+ source_file_dependencies:
3274
+ - vllm/
3275
+ - tests/entrypoints/llm
3276
+ commands:
3277
+ - export VLLM_WORKER_MULTIPROC_METHOD=spawn
3278
+ - pytest -v -s entrypoints/llm --ignore=entrypoints/llm/test_generate.py --ignore=entrypoints/llm/test_collective_rpc.py --ignore=entrypoints/llm/offline_mode
3279
+ - pytest -v -s entrypoints/llm/test_generate.py # it needs a clean process
3280
+ - pytest -v -s entrypoints/llm/offline_mode # Needs to avoid interference with other tests
3281
+
3282
+- label: ":amd: (MI355 DPX) Entrypoints Integration (API Server) %N"
3283
timeout_in_minutes: 65
3284
parallelism: 4
3285
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
3286
dind: false
3862
- agent_pool: mi355_1
3287
+ agent_pool: mi355_dpx
3288
optional: true
3289
fast_check: true
3290
working_dir: "/vllm-workspace/tests"
@@ -3873,11 +3298,11 @@ steps:
3298
- if [ "$$BUILDKITE_PARALLEL_JOB" = "1" ]; then PYTHONPATH=/vllm-workspace pytest -v -s entrypoints/serve/dev/rpc; fi
3299
- pytest -v -s entrypoints/scale_out --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
3300
3876
-- label: ":amd: (MI355) Entrypoints Integration (OpenAI API completion)" # TBD
3301
+- label: ":amd: (MI355 DPX) Entrypoints Integration (OpenAI API completion)" # TBD
3302
timeout_in_minutes: 65
3303
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
3304
dind: false
3880
- agent_pool: mi355_1
3305
+ agent_pool: mi355_dpx
3306
fast_check: true
3307
optional: true
3308
working_dir: "/vllm-workspace/tests"
@@ -3889,11 +3314,11 @@ steps:
3314
- pytest -v -s entrypoints/openai/completion --ignore=entrypoints/openai/completion/test_tensorizer_entrypoint.py
3315
- pytest -v -s entrypoints/openai --ignore=entrypoints/openai/completion --ignore=entrypoints/openai/chat_completion --ignore=entrypoints/openai/responses --ignore=entrypoints/openai/correctness
3316
3892
-- label: ":amd: (MI355) Entrypoints Integration (OpenAI API chat_completion)" # TBD
3317
+- label: ":amd: (MI355 DPX) Entrypoints Integration (OpenAI API chat_completion)" # TBD
3318
timeout_in_minutes: 70
3319
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
3320
dind: false
3896
- agent_pool: mi355_1
3321
+ agent_pool: mi355_dpx
3322
fast_check: true
3323
optional: true
3324
working_dir: "/vllm-workspace/tests"
@@ -3904,11 +3329,103 @@ steps:
3329
- export VLLM_WORKER_MULTIPROC_METHOD=spawn
3330
- pytest -v -s entrypoints/openai/chat_completion
3331
3907
-- label: ":amd: (MI355) Entrypoints Integration (API Server Generate)" # TBD
3332
+- label: ":amd: (MI355 DPX) Entrypoints Integration (API Server Generate)" # TBD
3333
timeout_in_minutes: 65
3334
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
3335
dind: false
3911
- agent_pool: mi355_1
3336
+ agent_pool: mi355_dpx
3337
+ optional: true
3338
+ fast_check: true
3339
+ working_dir: "/vllm-workspace/tests"
3340
+ source_file_dependencies:
3341
+ - vllm/
3342
+ - tests/tool_use
3343
+ - tests/entrypoints/tool_parsers
3344
+ - tests/entrypoints/anthropic
3345
+ - tests/entrypoints/cohere
3346
+ - tests/entrypoints/generate
3347
+ commands:
3348
+ - pytest -v -s tool_use
3349
+ - pytest -v -s entrypoints/tool_parsers
3350
+ - pytest -v -s entrypoints/generate
3351
+ - pytest -v -s entrypoints/anthropic
3352
+ - pytest -v -s entrypoints/cohere
3353
+
3354
+- label: ":amd: (MI355 DPX) OpenAI API Correctness" # TBD
3355
+ timeout_in_minutes: 30
3356
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
3357
+ dind: false
3358
+ agent_pool: mi355_dpx
3359
+ optional: true
3360
+ working_dir: "/vllm-workspace/tests"
3361
+ source_file_dependencies:
3362
+ - csrc/
3363
+ - vllm/entrypoints/openai/
3364
+ - vllm/model_executor/layers/
3365
+ - vllm/v1/attention/backends/
3366
+ - vllm/v1/attention/selector.py
3367
+ - vllm/_aiter_ops.py
3368
+ - vllm/platforms/rocm.py
3369
+ - vllm/model_executor/model_loader/
3370
+ commands:
3371
+ - bash ../tools/install_torchcodec_rocm.sh || exit 1
3372
+ - pytest -s entrypoints/openai/correctness/
3373
+
3374
+- label: ":amd: (MI355 DPX) Entrypoints Integration (API Server) %N (MI355 suite)"
3375
+ timeout_in_minutes: 65
3376
+ parallelism: 4
3377
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
3378
+ dind: false
3379
+ agent_pool: mi355_dpx
3380
+ optional: true
3381
+ fast_check: true
3382
+ working_dir: "/vllm-workspace/tests"
3383
+ source_file_dependencies:
3384
+ - vllm/
3385
+ - tests/entrypoints/serve
3386
+ - tests/entrypoints/scale_out
3387
+ commands:
3388
+ - export VLLM_WORKER_MULTIPROC_METHOD=spawn
3389
+ - pytest -v -s entrypoints/serve --ignore=entrypoints/serve/dev/rpc --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
3390
+ - if [ "$$BUILDKITE_PARALLEL_JOB" = "1" ]; then PYTHONPATH=/vllm-workspace pytest -v -s entrypoints/serve/dev/rpc; fi
3391
+ - pytest -v -s entrypoints/scale_out --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
3392
+
3393
+- label: ":amd: (MI355 DPX) Entrypoints Integration (OpenAI API completion) (MI355 suite)" # TBD
3394
+ timeout_in_minutes: 65
3395
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
3396
+ dind: false
3397
+ agent_pool: mi355_dpx
3398
+ fast_check: true
3399
+ optional: true
3400
+ working_dir: "/vllm-workspace/tests"
3401
+ source_file_dependencies:
3402
+ - vllm/
3403
+ - tests/entrypoints/openai
3404
+ commands:
3405
+ - export VLLM_WORKER_MULTIPROC_METHOD=spawn
3406
+ - pytest -v -s entrypoints/openai/completion --ignore=entrypoints/openai/completion/test_tensorizer_entrypoint.py
3407
+ - pytest -v -s entrypoints/openai --ignore=entrypoints/openai/completion --ignore=entrypoints/openai/chat_completion --ignore=entrypoints/openai/responses --ignore=entrypoints/openai/correctness
3408
+
3409
+- label: ":amd: (MI355 DPX) Entrypoints Integration (OpenAI API chat_completion) (MI355 suite)" # TBD
3410
+ timeout_in_minutes: 70
3411
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
3412
+ dind: false
3413
+ agent_pool: mi355_dpx
3414
+ fast_check: true
3415
+ optional: true
3416
+ working_dir: "/vllm-workspace/tests"
3417
+ source_file_dependencies:
3418
+ - vllm/
3419
+ - tests/entrypoints/openai
3420
+ commands:
3421
+ - export VLLM_WORKER_MULTIPROC_METHOD=spawn
3422
+ - pytest -v -s entrypoints/openai/chat_completion
3423
+
3424
+- label: ":amd: (MI355 DPX) Entrypoints Integration (API Server Generate) (MI355 suite)" # TBD
3425
+ timeout_in_minutes: 65
3426
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
3427
+ dind: false
3428
+ agent_pool: mi355_dpx
3429
fast_check: true
3430
optional: true
3431
working_dir: "/vllm-workspace/tests"
@@ -3926,11 +3443,11 @@ steps:
3443
- pytest -v -s entrypoints/anthropic
3444
- pytest -v -s entrypoints/cohere
3445
3929
-- label: ":amd: (MI355) Entrypoints Integration (Speech to Text)" # TBD
3446
+- label: ":amd: (MI355 DPX) Entrypoints Integration (Speech to Text)" # TBD
3447
timeout_in_minutes: 60
3931
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi355]
3448
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
3449
dind: false
3933
- agent_pool: mi355_1
3450
+ agent_pool: mi355_dpx
3451
fast_check: true
3452
working_dir: "/vllm-workspace/tests"
3453
source_file_dependencies:
@@ -3941,11 +3458,11 @@ steps:
3458
- export VLLM_WORKER_MULTIPROC_METHOD=spawn
3459
- pytest -v -s entrypoints/speech_to_text
3460
3944
-- label: ":amd: (MI355) Entrypoints Integration (Multimodal)"
3461
+- label: ":amd: (MI355 DPX) Entrypoints Integration (Multimodal)"
3462
timeout_in_minutes: 55
3946
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi355]
3463
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
3464
dind: false
3948
- agent_pool: mi355_1
3465
+ agent_pool: mi355_dpx
3466
fast_check: true
3467
working_dir: "/vllm-workspace/tests"
3468
source_file_dependencies:
@@ -3956,11 +3473,11 @@ steps:
3473
- export VLLM_WORKER_MULTIPROC_METHOD=spawn
3474
- pytest -v -s entrypoints/multimodal
3475
3959
-- label: ":amd: (MI355) Entrypoints Integration (Pooling) Shard %N" # TBD
3476
+- label: ":amd: (MI355 DPX) Entrypoints Integration (Pooling) Shard %N" # TBD
3477
timeout_in_minutes: 65
3478
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
3479
dind: false
3963
- agent_pool: mi355_1
3480
+ agent_pool: mi355_dpx
3481
fast_check: true
3482
parallelism: 4
3483
working_dir: "/vllm-workspace/tests"
@@ -3974,6 +3491,22 @@ steps:
3491
3492
#----------------------------------------------------------- mi355 · evals -----------------------------------------------------------#
3493
3494
+- label: ":amd: (MI355 DPX) Multimodal Accuracy Eval (Small Models)" # TBD
3495
+ timeout_in_minutes: 35
3496
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
3497
+ dind: false
3498
+ agent_pool: mi355_dpx
3499
+ optional: true
3500
+ working_dir: "/vllm-workspace/.buildkite/lm-eval-harness"
3501
+ source_file_dependencies:
3502
+ - vllm/multimodal/
3503
+ - vllm/inputs/
3504
+ - vllm/v1/core/
3505
+ - vllm/platforms/rocm.py
3506
+ - vllm/model_executor/model_loader/
3507
+ commands:
3508
+ - pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-mm-small.txt --tp-size=1
3509
+
3510
- label: ":amd: (MI355) GPQA Eval (GPT-OSS)" # TBD
3511
timeout_in_minutes: 40
3512
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
@@ -3997,11 +3530,11 @@ steps:
3530
- uv pip install --system 'gpt-oss[eval]==0.0.5'
3531
- pytest -s -v evals/gpt_oss/test_gpqa_correctness.py --config-list-file=configs/models-gfx950.txt
3532
4000
-- label: ":amd: (MI355) LM Eval Watermarking"
3533
+- label: ":amd: (MI355 DPX) LM Eval Watermarking"
3534
timeout_in_minutes: 45
3535
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
3536
dind: false
4004
- agent_pool: mi355_1
3537
+ agent_pool: mi355_dpx
3538
num_gpus: 1
3539
optional: true
3540
working_dir: "/vllm-workspace/tests"
@@ -4298,6 +3831,41 @@ steps:
3831
3832
#--------------------------------------------------------- mi355 · examples ----------------------------------------------------------#
3833
3834
+- label: ":amd: (MI355 DPX) Examples" # TBD
3835
+ timeout_in_minutes: 75
3836
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
3837
+ dind: false
3838
+ agent_pool: mi355_dpx
3839
+ optional: true
3840
+ working_dir: "/vllm-workspace/examples"
3841
+ source_file_dependencies:
3842
+ - vllm/entrypoints
3843
+ - vllm/multimodal
3844
+ - examples/
3845
+ - vllm/platforms/rocm.py
3846
+ commands:
3847
+ # Basic
3848
+ # Multi-modal models
3849
+ # Pooling models
3850
+ # Features demo
3851
+ - pip install --no-deps tensorizer
3852
+ - python3 basic/offline_inference/chat.py
3853
+ - python3 basic/offline_inference/generate.py --model facebook/opt-125m
3854
+ - python3 basic/offline_inference/generate.py --model meta-llama/Llama-2-13b-chat-hf --cpu-offload-gb 10
3855
+ - python3 basic/offline_inference/classify.py
3856
+ - python3 basic/offline_inference/embed.py
3857
+ - python3 basic/offline_inference/score.py
3858
+ - python3 generate/multimodal/audio_language_offline.py --seed 0
3859
+ - python3 generate/multimodal/vision_language_offline.py --seed 0
3860
+ - python3 generate/multimodal/vision_language_multi_image_offline.py --seed 0
3861
+ - python3 generate/multimodal/encoder_decoder_multimodal_offline.py --model-type whisper --seed 0
3862
+ - python3 pooling/embed/vision_embedding_offline.py --seed 0
3863
+ - python3 features/automatic_prefix_caching/prefix_caching_offline.py
3864
+ - python3 deployment/llm_engine_example.py
3865
+ - python3 features/tensorize_vllm_model.py --model facebook/opt-125m serialize --serialized-directory /tmp/ --suffix v1 && python3 features/tensorize_vllm_model.py --model facebook/opt-125m deserialize --path-to-tensors /tmp/vllm/facebook/opt-125m/v1/model.tensors
3866
+ - python3 features/speculative_decoding/spec_decode_offline.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
3867
+ - 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
3868
+
3869
- label: ":amd: (MI355) Examples" # TBD
3870
timeout_in_minutes: 75
3871
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
@@ -4335,11 +3903,11 @@ steps:
3903
3904
#---------------------------------------------------------- mi355 · kernels ----------------------------------------------------------#
3905
4338
-- label: ":amd: (MI355) Attention DiffKV Kernels"
3906
+- label: ":amd: (MI355 DPX) Attention DiffKV Kernels"
3907
timeout_in_minutes: 35
3908
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
3909
dind: false
4342
- agent_pool: mi355_1
3910
+ agent_pool: mi355_dpx
3911
num_gpus: 1
3912
optional: true
3913
working_dir: "/vllm-workspace/tests"
@@ -4352,29 +3920,6 @@ steps:
3920
commands:
3921
- pytest -v -s kernels/attention/test_triton_unified_attention_diffkv.py
3922
4355
-- label: ":amd: (MI355) Core Operation Kernels Shard %N" # TBD
4356
- timeout_in_minutes: 40
4357
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4358
- dind: false
4359
- agent_pool: mi355_1
4360
- parallelism: 3
4361
- working_dir: "/vllm-workspace/tests"
4362
- source_file_dependencies:
4363
- - csrc/
4364
- - tests/kernels/core
4365
- - tests/kernels/test_top_k_per_row.py
4366
- - tests/kernels/test_concat_mla_q.py
4367
- - tests/kernels/test_rocm_fp32_router_gemm.py
4368
- - vllm/model_executor/layers/fused_moe/router/gate_linear.py
4369
- - vllm/model_executor/layers/fused_moe/router/rocm_fp32_router_gemm.py
4370
- - vllm/model_executor/layers/rotary_embedding/
4371
- - vllm/_aiter_ops.py
4372
- - vllm/platforms/rocm.py
4373
- - tests/kernels/test_fused_qk_norm_rope_gate.py
4374
- optional: true
4375
- commands:
4376
- - pytest -v -s kernels/core --ignore=kernels/core/test_minimax_reduce_rms.py kernels/test_fused_qk_norm_rope_gate.py kernels/test_concat_mla_q.py kernels/test_top_k_per_row.py kernels/test_rocm_fp32_router_gemm.py --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
4377
-
3923
- label: ":amd: (MI355) MiniMax Reduce RMS Kernels" # TBD
3924
timeout_in_minutes: 45
3925
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
@@ -4399,11 +3944,11 @@ steps:
3944
commands:
3945
- pytest -v -s kernels/core/test_minimax_reduce_rms.py
3946
4402
-- label: ":amd: (MI355) Kernels" # TBD
3947
+- label: ":amd: (MI355 DPX) Kernels" # TBD
3948
timeout_in_minutes: 25
3949
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
3950
dind: false
4406
- agent_pool: mi355_1
3951
+ agent_pool: mi355_dpx
3952
working_dir: "/vllm-workspace/"
3953
source_file_dependencies:
3954
- csrc/quantization/fp4/
@@ -4427,11 +3972,11 @@ steps:
3972
- python3 examples/basic/offline_inference/chat.py --attention-backend TRITON_ATTN
3973
- pytest -v -s tests/kernels/attention/test_attention_selector.py
3974
4430
-- label: ":amd: (MI355) MLA Kernels"
3975
+- label: ":amd: (MI355 DPX) MLA Kernels"
3976
timeout_in_minutes: 40
3977
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
3978
dind: false
4434
- agent_pool: mi355_1
3979
+ agent_pool: mi355_dpx
3980
num_gpus: 1
3981
optional: true
3982
working_dir: "/vllm-workspace/"
@@ -4451,11 +3996,11 @@ steps:
3996
- pytest -v -s tests/kernels/attention/test_rocm_aiter_mla_fp8_support.py
3997
- pytest -v -s tests/kernels/attention/test_rocm_aiter_mla_op_registration.py
3998
4454
-- label: ":amd: (MI355) DeepSeek V4 Fused Kernels"
3999
+- label: ":amd: (MI355 DPX) DeepSeek V4 Fused Kernels"
4000
timeout_in_minutes: 60
4001
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4002
dind: false
4458
- agent_pool: mi355_1
4003
+ agent_pool: mi355_dpx
4004
num_gpus: 1
4005
optional: true
4006
working_dir: "/vllm-workspace/tests"
@@ -4467,11 +4012,11 @@ steps:
4012
commands:
4013
- pytest -v -s kernels/test_fused_deepseek_v4_qnorm_rope_kv_insert.py
4014
4470
-- label: ":amd: (MI355) Attention Kernels Shard %N" # TBD
4015
+- label: ":amd: (MI355 DPX) Attention Kernels Shard %N" # TBD
4016
timeout_in_minutes: 90
4017
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4018
dind: false
4474
- agent_pool: mi355_1
4019
+ agent_pool: mi355_dpx
4020
parallelism: 2
4021
working_dir: "/vllm-workspace/tests"
4022
source_file_dependencies:
@@ -4487,11 +4032,11 @@ steps:
4032
commands:
4033
- pytest -v -s kernels/attention --ignore=kernels/attention/test_triton_unified_attention_diffkv.py --ignore=kernels/attention/test_rocm_aiter_mla_decode.py --ignore=kernels/attention/test_rocm_aiter_mla_decode_metadata.py --ignore=kernels/attention/test_rocm_aiter_mla_fp8_support.py --ignore=kernels/attention/test_rocm_aiter_mla_op_registration.py --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
4034
4490
-- label: ":amd: (MI355) MoE Kernels Shard %N" # TBD
4035
+- label: ":amd: (MI355 DPX) MoE Kernels Shard %N (MI355 suite)" # TBD
4036
timeout_in_minutes: 55
4037
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4038
dind: false
4494
- agent_pool: mi355_1
4039
+ agent_pool: mi355_dpx
4040
parallelism: 5
4041
working_dir: "/vllm-workspace/tests"
4042
source_file_dependencies:
@@ -4631,11 +4176,11 @@ steps:
4176
- amd-smi
4177
- pytest -v -s kernels/quantization/test_nvfp4_emulation.py -k test_nvfp4_dense_emulation --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
4178
4634
-- label: ":amd: (MI355) FP8 MoE Kernels" # TBD
4179
+- label: ":amd: (MI355 DPX) FP8 MoE Kernels" # TBD
4180
timeout_in_minutes: 40
4181
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4182
dind: false
4638
- agent_pool: mi355_1
4183
+ agent_pool: mi355_dpx
4184
working_dir: "/vllm-workspace/tests"
4185
source_file_dependencies:
4186
- csrc/moe/
@@ -4678,13 +4223,50 @@ steps:
4223
commands:
4224
- pytest -v -s kernels/moe/test_deepep_moe.py
4225
4226
+#----------------------------------------------------------- mi355 · lora ------------------------------------------------------------#
4227
+
4228
+- label: ":amd: (MI355 DPX) LoRA Shard %N" # TBD
4229
+ timeout_in_minutes: 85
4230
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4231
+ dind: false
4232
+ agent_pool: mi355_dpx
4233
+ parallelism: 4
4234
+ optional: true
4235
+ working_dir: "/vllm-workspace/tests"
4236
+ source_file_dependencies:
4237
+ - vllm/lora
4238
+ - tests/lora
4239
+ - vllm/platforms/rocm.py
4240
+ commands:
4241
+ - pytest -v -s lora --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --ignore=lora/test_chatglm3_tp.py --ignore=lora/test_llama_tp.py --ignore=lora/test_qwen3_with_multi_loras.py --ignore=lora/test_olmoe_tp.py --ignore=lora/test_deepseekv2_tp.py --ignore=lora/test_gptoss_tp.py --ignore=lora/test_qwen3moe_tp.py --ignore=lora/test_qwen35_densemodel_lora.py
4242
+
4243
#------------------------------------------------------ mi355 · models / basic -------------------------------------------------------#
4244
4683
-- label: ":amd: (MI355) GLM5Next"
4245
+- label: ":amd: (MI355 DPX) Basic Models (Other)" # TBD
4246
+ timeout_in_minutes: 65
4247
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4248
+ dind: false
4249
+ agent_pool: mi355_dpx
4250
+ optional: true
4251
+ working_dir: "/vllm-workspace/tests"
4252
+ source_file_dependencies:
4253
+ - vllm/
4254
+ - tests/models/test_terratorch.py
4255
+ - tests/models/transformers/test_backend.py
4256
+ - tests/models/test_registry.py
4257
+ - tests/models/test_hyv4_rocm.py
4258
+ - tests/models/test_vision.py
4259
+ - tests/models/test_deepseek_v4_vl_rocm.py
4260
+ commands:
4261
+ - pytest -v -s models/test_terratorch.py models/transformers/test_backend.py models/test_registry.py models/test_deepseek_v4_vl_rocm.py
4262
+ - VLLM_ROCM_USE_AITER=1 pytest -v -s models/test_hyv4_rocm.py -m 'not distributed'
4263
+ - pytest -v -s models/test_vision.py::test_simple_mrope_vision_model_spatial_merge
4264
+
4265
+- label: ":amd: (MI355 DPX) GLM5Next"
4266
timeout_in_minutes: 35
4267
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4268
dind: false
4687
- agent_pool: mi355_1
4269
+ agent_pool: mi355_dpx
4270
num_gpus: 1
4271
optional: true
4272
working_dir: "/vllm-workspace/tests"
@@ -4695,11 +4277,11 @@ steps:
4277
commands:
4278
- pytest -v -s models/glm5next
4279
4698
-- label: ":amd: (MI355) Basic Models (Extra Initialization) Shard %N" # TBD
4280
+- label: ":amd: (MI355 DPX) Basic Models (Extra Initialization) Shard %N" # TBD
4281
timeout_in_minutes: 60
4282
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4283
dind: false
4702
- agent_pool: mi355_1
4284
+ agent_pool: mi355_dpx
4285
parallelism: 6
4286
working_dir: "/vllm-workspace/tests"
4287
source_file_dependencies:
@@ -4715,30 +4297,67 @@ steps:
4297
4298
#----------------------------------------------------- mi355 · models / language -----------------------------------------------------#
4299
4718
-- label: ":amd: (MI355) Language Models (Hybrid) Shard %N" # TBD
4719
- timeout_in_minutes: 65
4300
+- label: ":amd: (MI355 DPX) Language Models (Standard)" # TBD
4301
+ timeout_in_minutes: 60
4302
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4303
dind: false
4722
- agent_pool: mi355_1
4723
- parallelism: 2
4304
+ agent_pool: mi355_dpx
4305
optional: true
4306
working_dir: "/vllm-workspace/tests"
4307
source_file_dependencies:
4308
- vllm/
4728
- - tests/models/language/generation
4309
+ - tests/models/language
4310
commands:
4730
- - export GIT_CONFIG_COUNT=1
4731
- - export GIT_CONFIG_KEY_0=http.version
4732
- - export GIT_CONFIG_VALUE_0=HTTP/1.1
4733
- - MAMBA_FORCE_BUILD=TRUE uv pip install --system --no-build-isolation 'git+https://github.com/AndreasKaratzas/mamba@fix-rocm-7.0-warp-size-constexpr'
4311
+ - pip freeze | grep -E 'torch'
4312
+ - pytest -v -s models/language -m 'core_model and (not slow_test)'
4313
+
4314
+- label: ":amd: (MI355 DPX) Language Models (Extra Standard) Shard %N" # TBD
4315
+ timeout_in_minutes: 40
4316
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4317
+ dind: false
4318
+ agent_pool: mi355_dpx
4319
+ optional: true
4320
+ parallelism: 2
4321
+ working_dir: "/vllm-workspace/tests"
4322
+ source_file_dependencies:
4323
+ - vllm/model_executor/models/
4324
+ - vllm/model_executor/model_loader/
4325
+ - vllm/model_executor/layers/
4326
+ - vllm/v1/attention/backends/
4327
+ - vllm/v1/attention/selector.py
4328
+ - tests/models/language/pooling/test_embedding.py
4329
+ - tests/models/language/generation/test_common.py
4330
+ - tests/models/language/pooling/test_classification.py
4331
+ - vllm/_aiter_ops.py
4332
+ - vllm/platforms/rocm.py
4333
+ commands:
4334
+ - pip freeze | grep -E 'torch'
4335
+ - pytest -v -s models/language -m 'core_model and slow_test' --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
4336
+
4337
+- label: ":amd: (MI355 DPX) Language Models (Hybrid) Shard %N" # TBD
4338
+ timeout_in_minutes: 65
4339
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4340
+ dind: false
4341
+ agent_pool: mi355_dpx
4342
+ parallelism: 2
4343
+ optional: true
4344
+ working_dir: "/vllm-workspace/tests"
4345
+ source_file_dependencies:
4346
+ - vllm/
4347
+ - tests/models/language/generation
4348
+ commands:
4349
+ - export GIT_CONFIG_COUNT=1
4350
+ - export GIT_CONFIG_KEY_0=http.version
4351
+ - export GIT_CONFIG_VALUE_0=HTTP/1.1
4352
+ - MAMBA_FORCE_BUILD=TRUE uv pip install --system --no-build-isolation 'git+https://github.com/AndreasKaratzas/mamba@fix-rocm-7.0-warp-size-constexpr'
4353
- CAUSAL_CONV1D_FORCE_BUILD=TRUE uv pip install --system --no-build-isolation 'git+https://github.com/Dao-AILab/causal-conv1d@v1.6.0'
4354
- pytest -v -s models/language/generation -m hybrid_model -k 'not granite-4.0-tiny-preview' --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
4355
4737
-- label: ":amd: (MI355) Granite Language Model Compatibility"
4356
+- label: ":amd: (MI355 DPX) Granite Language Model Compatibility"
4357
timeout_in_minutes: 80
4358
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4359
dind: false
4741
- agent_pool: mi355_1
4360
+ agent_pool: mi355_dpx
4361
num_gpus: 1
4362
optional: true
4363
working_dir: "/vllm-workspace/tests"
@@ -4754,11 +4373,11 @@ steps:
4373
- CAUSAL_CONV1D_FORCE_BUILD=TRUE uv pip install --system --no-build-isolation 'git+https://github.com/Dao-AILab/causal-conv1d@v1.6.0'
4374
- pytest -v -s models/language/generation -m hybrid_model -k 'granite-4.0-tiny-preview'
4375
4757
-- label: ":amd: (MI355) Language Models (Extended Generation)" # TBD
4376
+- label: ":amd: (MI355 DPX) Language Models (Extended Generation)" # TBD
4377
timeout_in_minutes: 95
4378
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4379
dind: false
4761
- agent_pool: mi355_1
4380
+ agent_pool: mi355_dpx
4381
working_dir: "/vllm-workspace/tests"
4382
source_file_dependencies:
4383
- vllm/
@@ -4769,12 +4388,12 @@ steps:
4388
- CAUSAL_CONV1D_FORCE_BUILD=TRUE uv pip install --system --no-build-isolation 'git+https://github.com/Dao-AILab/causal-conv1d@v1.6.0'
4389
- pytest -v -s models/language/generation -m '(not core_model) and (not hybrid_model)'
4390
4772
-- label: ":amd: (MI355) Language Models (Extended Pooling) Shard %N"
4391
+- label: ":amd: (MI355 DPX) Language Models (Extended Pooling) Shard %N"
4392
timeout_in_minutes: 95
4393
parallelism: 4
4394
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4395
dind: false
4777
- agent_pool: mi355_1
4396
+ agent_pool: mi355_dpx
4397
optional: true
4398
working_dir: "/vllm-workspace/tests"
4399
source_file_dependencies:
@@ -4783,11 +4402,11 @@ steps:
4402
commands:
4403
- pytest -v -s models/language/pooling -m 'not core_model' --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
4404
4786
-- label: ":amd: (MI355) Language Models (PPL)" # TBD
4405
+- label: ":amd: (MI355 DPX) Language Models (PPL)" # TBD
4406
timeout_in_minutes: 65
4407
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4408
dind: false
4790
- agent_pool: mi355_1
4409
+ agent_pool: mi355_dpx
4410
optional: true
4411
working_dir: "/vllm-workspace/tests"
4412
source_file_dependencies:
@@ -4814,11 +4433,11 @@ steps:
4433
commands:
4434
- pytest -v -s models/language/generation_ppl_test
4435
4817
-- label: ":amd: (MI355) Language Models (Standard)" # TBD
4436
+- label: ":amd: (MI355 DPX) Language Models (Standard) (MI355 suite)" # TBD
4437
timeout_in_minutes: 60
4438
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4439
dind: false
4821
- agent_pool: mi355_1
4440
+ agent_pool: mi355_dpx
4441
working_dir: "/vllm-workspace/tests"
4442
source_file_dependencies:
4443
- vllm/
@@ -4830,11 +4449,40 @@ steps:
4449
4450
#---------------------------------------------------- mi355 · models / multimodal ----------------------------------------------------#
4451
4833
-- label: ":amd: (MI355) Multimodal Models (Extended Generation 1)" # TBD
4452
+- label: ":amd: (MI355 DPX) Multimodal Models (Standard) 1: qwen2" # TBD
4453
+ timeout_in_minutes: 65
4454
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4455
+ dind: false
4456
+ agent_pool: mi355_dpx
4457
+ optional: true
4458
+ working_dir: "/vllm-workspace/tests"
4459
+ source_file_dependencies:
4460
+ - vllm/
4461
+ - tests/models/multimodal
4462
+ commands:
4463
+ - pytest -v -s models/multimodal/generation/test_common.py -m core_model -k "qwen2"
4464
+ - pytest -v -s models/multimodal/generation/test_ultravox.py -m core_model
4465
+
4466
+- label: ":amd: (MI355 DPX) Multimodal Models (Standard) 2: qwen3 + gemma" # TBD
4467
+ timeout_in_minutes: 55
4468
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4469
+ dind: false
4470
+ agent_pool: mi355_dpx
4471
+ optional: true
4472
+ working_dir: "/vllm-workspace/tests"
4473
+ source_file_dependencies:
4474
+ - vllm/
4475
+ - tests/models/multimodal
4476
+ commands:
4477
+ - pytest -v -s models/multimodal/generation/test_common.py -m core_model -k "qwen3 or gemma"
4478
+ - pytest -v -s models/multimodal/generation/test_mm_prefix_lm.py -m core_model
4479
+ - pytest -v -s models/multimodal/generation/test_qwen2_5_vl.py -m core_model
4480
+
4481
+- label: ":amd: (MI355 DPX) Multimodal Models (Extended Generation 1)" # TBD
4482
timeout_in_minutes: 90
4483
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4484
dind: false
4837
- agent_pool: mi355_1
4485
+ agent_pool: mi355_dpx
4486
optional: true
4487
working_dir: "/vllm-workspace/tests"
4488
source_file_dependencies:
@@ -4842,166 +4490,518 @@ steps:
4490
- tests/models/multimodal/generation
4491
- tests/models/multimodal/test_mapping.py
4492
commands:
4845
- - pytest -v -s models/multimodal/generation -m 'not core_model' --ignore models/multimodal/generation/test_common.py
4846
- - pytest -v -s models/multimodal/test_mapping.py
4847
-
4848
-- label: ":amd: (MI355) Multimodal Models (Extended Generation 3)" # TBD
4849
- timeout_in_minutes: 90
4850
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4851
- dind: false
4852
- agent_pool: mi355_1
4493
+ - pytest -v -s models/multimodal/generation -m 'not core_model' --ignore models/multimodal/generation/test_common.py
4494
+ - pytest -v -s models/multimodal/test_mapping.py
4495
+
4496
+- label: ":amd: (MI355 DPX) Multimodal Models (Extended Generation 3)" # TBD
4497
+ timeout_in_minutes: 90
4498
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4499
+ dind: false
4500
+ agent_pool: mi355_dpx
4501
+ optional: true
4502
+ working_dir: "/vllm-workspace/tests"
4503
+ source_file_dependencies:
4504
+ - vllm/
4505
+ - tests/models/multimodal/generation
4506
+ commands:
4507
+ - pytest -v -s models/multimodal/generation/test_common.py -m 'split(group=1) and not core_model'
4508
+
4509
+- label: ":amd: (MI355 DPX) Multimodal Models (Extended Pooling)" # TBD
4510
+ timeout_in_minutes: 75
4511
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4512
+ dind: false
4513
+ agent_pool: mi355_dpx
4514
+ optional: true
4515
+ working_dir: "/vllm-workspace/tests"
4516
+ source_file_dependencies:
4517
+ - vllm/
4518
+ - tests/models/multimodal/pooling
4519
+ commands:
4520
+ - pytest -v -s models/multimodal/pooling -m 'not core_model'
4521
+
4522
+- label: ":amd: (MI355 DPX) Multimodal Models (Standard) 1: qwen2 (MI355 suite)" # TBD
4523
+ timeout_in_minutes: 65
4524
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4525
+ dind: false
4526
+ agent_pool: mi355_dpx
4527
+ optional: true
4528
+ working_dir: "/vllm-workspace/tests"
4529
+ source_file_dependencies:
4530
+ - vllm/
4531
+ - tests/models/multimodal
4532
+ commands:
4533
+ - pytest -v -s models/multimodal/generation/test_common.py -m core_model -k "qwen2"
4534
+ - pytest -v -s models/multimodal/generation/test_ultravox.py -m core_model
4535
+
4536
+- label: ":amd: (MI355 DPX) Multimodal Models (Standard) 4: other + whisper" # TBD
4537
+ timeout_in_minutes: 70
4538
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4539
+ dind: false
4540
+ agent_pool: mi355_dpx
4541
+ optional: true
4542
+ working_dir: "/vllm-workspace/tests"
4543
+ source_file_dependencies:
4544
+ - vllm/
4545
+ - tests/models/multimodal
4546
+ commands:
4547
+ - pytest -v -s models/multimodal -m core_model --ignore models/multimodal/generation/test_common.py --ignore models/multimodal/generation/test_ultravox.py --ignore models/multimodal/generation/test_qwen2_5_vl.py --ignore models/multimodal/generation/test_qwen2_vl.py --ignore models/multimodal/generation/test_whisper.py --ignore models/multimodal/generation/test_mm_prefix_lm.py --ignore models/multimodal/generation/test_memory_leak.py --ignore models/multimodal/generation/test_vit_cudagraph.py --ignore models/multimodal/processing
4548
+ - pytest -v -s models/multimodal/generation/test_vit_cudagraph.py -m core_model
4549
+ - pytest models/multimodal/generation/test_memory_leak.py -m core_model
4550
+ - cd .. && VLLM_WORKER_MULTIPROC_METHOD=spawn pytest -v -s tests/models/multimodal/generation/test_whisper.py -m core_model
4551
+
4552
+#----------------------------------------------------- mi355 · models / quantized -----------------------------------------------------#
4553
+
4554
+- label: ":amd: (MI355 DPX) Quantized Models" # TBD
4555
+ timeout_in_minutes: 80
4556
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4557
+ dind: false
4558
+ agent_pool: mi355_dpx
4559
+ working_dir: "/vllm-workspace/tests"
4560
+ source_file_dependencies:
4561
+ - vllm/model_executor/layers/quantization
4562
+ - tests/models/quantization
4563
+ - vllm/_aiter_ops.py
4564
+ - vllm/platforms/rocm.py
4565
+ - vllm/model_executor/model_loader/
4566
+ optional: true
4567
+ commands:
4568
+ - unset VLLM_USE_V2_MODEL_RUNNER
4569
+ - pytest -v -s models/quantization
4570
+
4571
+#------------------------------------------------------- mi355 · quantization --------------------------------------------------------#
4572
+
4573
+- label: ":amd: (MI355) Quantization Shard %N"
4574
+ timeout_in_minutes: 115
4575
+ parallelism: 4
4576
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4577
+ dind: false
4578
+ agent_pool: mi355_1
4579
+ working_dir: "/vllm-workspace/tests"
4580
+ source_file_dependencies:
4581
+ - csrc/
4582
+ - vllm/model_executor/layers/quantization
4583
+ - tests/quantization
4584
+ - vllm/_aiter_ops.py
4585
+ - vllm/platforms/rocm.py
4586
+ - vllm/model_executor/layers/fused_moe/oracle/unquantized.py
4587
+ - vllm/model_executor/layers/fused_moe/unquantized_fused_moe_method.py
4588
+ - tests/rocm/test_moe_weight_replay.py
4589
+ optional: true
4590
+ commands:
4591
+ - unset VLLM_USE_V2_MODEL_RUNNER
4592
+ - uv pip install --system torchao==0.17.0
4593
+ - uv pip install --system conch-triton-kernels
4594
+ - VLLM_TEST_FORCE_LOAD_FORMAT=auto pytest -v -s quantization/ --ignore quantization/test_blackwell_moe.py --ignore quantization/test_rocm_moe.py --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
4595
+ - if [ "$$BUILDKITE_PARALLEL_JOB" = "0" ]; then pytest -v -s rocm/test_moe_weight_replay.py; fi
4596
+
4597
+# - label: Quantized MoE Test (B200-MI355) # TBD
4598
+# timeout_in_minutes: 180
4599
+# mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4600
+# dind: false
4601
+# agent_pool: mi355_1
4602
+# working_dir: "/vllm-workspace/"
4603
+# source_file_dependencies:
4604
+# - tests/quantization/test_gfx950_moe.py
4605
+# - vllm/model_executor/models/deepseek_v2.py
4606
+# - vllm/model_executor/models/gpt_oss.py
4607
+# - vllm/model_executor/models/llama4.py
4608
+# - vllm/model_executor/layers/fused_moe
4609
+# - vllm/model_executor/layers/quantization/compressed_tensors
4610
+# - vllm/model_executor/layers/quantization/modelopt.py
4611
+# - vllm/model_executor/layers/quantization/mxfp4.py
4612
+# - vllm/v1/attention/backends/triton_attn.py
4613
+# - vllm/v1/attention/backends/rocm_attn.py
4614
+# - vllm/v1/attention/backends/rocm_aiter_fa.py
4615
+# - vllm/v1/attention/backends/mla/
4616
+# - vllm/v1/attention/selector.py
4617
+# - vllm/model_executor/layers/layernorm.py
4618
+# - vllm/_aiter_ops.py
4619
+# - vllm/platforms/rocm.py
4620
+# - vllm/model_executor/model_loader/
4621
+# commands:
4622
+# - pytest -s -v tests/quantization/test_gfx950_moe.py
4623
+
4624
+#--------------------------------------------------------- mi355 · samplers ----------------------------------------------------------#
4625
+
4626
+- label: ":amd: (MI355 DPX) Samplers Multimodal Beam Search"
4627
+ timeout_in_minutes: 45
4628
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4629
+ dind: false
4630
+ agent_pool: mi355_dpx
4631
+ num_gpus: 1
4632
+ optional: true
4633
+ working_dir: "/vllm-workspace/tests"
4634
+ source_file_dependencies:
4635
+ - vllm/model_executor/layers
4636
+ - vllm/sampling_metadata.py
4637
+ - vllm/v1/sample/
4638
+ - vllm/entrypoints/generate/beam_search/
4639
+ - tests/samplers
4640
+ - tests/conftest.py
4641
+ - vllm/_aiter_ops.py
4642
+ - vllm/platforms/rocm.py
4643
+ commands:
4644
+ - pytest -v -s samplers/test_beam_search.py::test_beam_search_passes_multimodal_data
4645
+
4646
+#------------------------------------------------------------ mi355 · v1 -------------------------------------------------------------#
4647
+
4648
+- label: ":amd: (MI355 DPX) Spec Decode Draft Model" # TBD
4649
+ timeout_in_minutes: 60
4650
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4651
+ dind: false
4652
+ agent_pool: mi355_dpx
4653
+ optional: true
4654
+ working_dir: "/vllm-workspace/tests"
4655
+ source_file_dependencies:
4656
+ - vllm/v1/spec_decode/
4657
+ - vllm/v1/worker/gpu/spec_decode/
4658
+ - vllm/model_executor/model_loader/
4659
+ - vllm/v1/sample/
4660
+ - vllm/model_executor/layers/
4661
+ - tests/v1/e2e/spec_decode/
4662
+ - vllm/platforms/rocm.py
4663
+ commands:
4664
+ - pytest -v -s v1/e2e/spec_decode/draft_model/
4665
+
4666
+- label: ":amd: (MI355 DPX) Spec Decode Eagle 1: DeepSeek + Qwen"
4667
+ timeout_in_minutes: 60
4668
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4669
+ dind: false
4670
+ agent_pool: mi355_dpx
4671
+ optional: true
4672
+ working_dir: "/vllm-workspace/tests"
4673
+ source_file_dependencies:
4674
+ - vllm/v1/spec_decode/
4675
+ - vllm/v1/worker/gpu/spec_decode/
4676
+ - vllm/model_executor/model_loader/
4677
+ - vllm/v1/sample/
4678
+ - vllm/model_executor/layers/
4679
+ - tests/v1/e2e/spec_decode/
4680
+ - vllm/platforms/rocm.py
4681
+ commands:
4682
+ - pytest -v -s v1/e2e/spec_decode/eagle/ -k "deepseek_eagle or qwen3_eagle3"
4683
+
4684
+- label: ":amd: (MI355 DPX) Spec Decode Eagle 2: Llama 3 + Qwen VL + Other"
4685
+ timeout_in_minutes: 60
4686
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4687
+ dind: false
4688
+ agent_pool: mi355_dpx
4689
+ optional: true
4690
+ working_dir: /vllm-workspace/tests
4691
+ source_file_dependencies:
4692
+ - vllm/v1/spec_decode/
4693
+ - vllm/v1/worker/gpu/spec_decode/
4694
+ - vllm/model_executor/model_loader/
4695
+ - vllm/v1/sample/
4696
+ - vllm/model_executor/layers/
4697
+ - tests/v1/e2e/spec_decode/
4698
+ - vllm/platforms/rocm.py
4699
+ commands:
4700
+ - pytest -v -s v1/e2e/spec_decode/eagle/ -k "not deepseek_eagle and not qwen3_eagle3"
4701
+
4702
+
4703
+- label: ":amd: (MI355 DPX) Spec Decode N-Gram + Suffix" # TBD
4704
+ timeout_in_minutes: 35
4705
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4706
+ dind: false
4707
+ agent_pool: mi355_dpx
4708
+ optional: true
4709
+ working_dir: "/vllm-workspace/tests"
4710
+ source_file_dependencies:
4711
+ - vllm/v1/spec_decode/
4712
+ - vllm/v1/worker/gpu/spec_decode/
4713
+ - vllm/model_executor/model_loader/
4714
+ - vllm/v1/sample/
4715
+ - vllm/model_executor/layers/
4716
+ - tests/v1/e2e/spec_decode/
4717
+ - tests/spec_decode/
4718
+ - vllm/platforms/rocm.py
4719
+ commands:
4720
+ - pytest -v -s v1/e2e/spec_decode/ngram_suffix/
4721
+ - python3 spec_decode/test_custom_proposer.py
4722
+
4723
+- label: ":amd: (MI355 DPX) V1 Core"
4724
+ timeout_in_minutes: 60
4725
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4726
+ dind: false
4727
+ agent_pool: mi355_dpx
4728
+ working_dir: "/vllm-workspace/tests"
4729
+ source_file_dependencies:
4730
+ - vllm/
4731
+ - tests/v1/core
4732
+ - tests/v1/executor
4733
+ - tests/v1/kv_offload
4734
+ - tests/v1/simple_kv_offload
4735
+ - tests/v1/worker
4736
+ - tests/v1/streaming_input
4737
+ - tests/v1/kv_connector/unit
4738
+ - tests/v1/ec_connector/unit
4739
+ - tests/v1/metrics
4740
+ - tests/entrypoints/openai/correctness/test_lmeval.py
4741
+ - vllm/config/
4742
+ - vllm/distributed/
4743
+ - vllm/engine/
4744
+ - vllm/entrypoints/pooling/
4745
+ - vllm/inputs/
4746
+ - vllm/lora/
4747
+ - vllm/model_executor/
4748
+ - vllm/multimodal/
4749
+ - vllm/outputs.py
4750
+ - vllm/platforms/
4751
+ - vllm/pooling_params.py
4752
+ - vllm/profiler/
4753
+ - vllm/sampling_params.py
4754
+ - vllm/tokenizers/
4755
+ - vllm/transformers_utils/
4756
+ - vllm/utils/
4757
+ - vllm/v1/
4758
+ optional: true
4759
+ commands:
4760
+ # - export HSA_NO_SCRATCH_RECLAIM=1
4761
+ - export VLLM_WORKER_MULTIPROC_METHOD=spawn
4762
+ - pytest -v -s -m 'not cpu_test' v1/core
4763
+
4764
+- label: ":amd: (MI355 DPX) V1 Executor + Worker"
4765
+ timeout_in_minutes: 60
4766
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4767
+ dind: false
4768
+ agent_pool: mi355_dpx
4769
+ working_dir: /vllm-workspace/tests
4770
+ source_file_dependencies:
4771
+ - vllm/
4772
+ - tests/v1/core
4773
+ - tests/v1/executor
4774
+ - tests/v1/kv_offload
4775
+ - tests/v1/simple_kv_offload
4776
+ - tests/v1/worker
4777
+ - tests/v1/streaming_input
4778
+ - tests/v1/kv_connector/unit
4779
+ - tests/v1/ec_connector/unit
4780
+ - tests/v1/metrics
4781
+ - tests/entrypoints/openai/correctness/test_lmeval.py
4782
+ - vllm/config/
4783
+ - vllm/distributed/
4784
+ - vllm/engine/
4785
+ - vllm/entrypoints/pooling/
4786
+ - vllm/inputs/
4787
+ - vllm/lora/
4788
+ - vllm/model_executor/
4789
+ - vllm/multimodal/
4790
+ - vllm/outputs.py
4791
+ - vllm/platforms/
4792
+ - vllm/pooling_params.py
4793
+ - vllm/profiler/
4794
+ - vllm/sampling_params.py
4795
+ - vllm/tokenizers/
4796
+ - vllm/transformers_utils/
4797
+ - vllm/utils/
4798
+ - vllm/v1/
4799
+ commands:
4800
+ - export VLLM_WORKER_MULTIPROC_METHOD=spawn
4801
+ - pytest -v -s v1/executor
4802
+ - pytest -v -s v1/worker
4803
+ - pytest -v -s v1/streaming_input
4804
+ optional: true
4805
+
4806
+- label: ":amd: (MI355 DPX) V1 KV Offload"
4807
+ timeout_in_minutes: 60
4808
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4809
+ dind: false
4810
+ agent_pool: mi355_dpx
4811
+ working_dir: /vllm-workspace/tests
4812
+ source_file_dependencies:
4813
+ - vllm/
4814
+ - tests/v1/core
4815
+ - tests/v1/executor
4816
+ - tests/v1/kv_offload
4817
+ - tests/v1/simple_kv_offload
4818
+ - tests/v1/worker
4819
+ - tests/v1/streaming_input
4820
+ - tests/v1/kv_connector/unit
4821
+ - tests/v1/ec_connector/unit
4822
+ - tests/v1/metrics
4823
+ - tests/entrypoints/openai/correctness/test_lmeval.py
4824
+ - vllm/config/
4825
+ - vllm/distributed/
4826
+ - vllm/engine/
4827
+ - vllm/entrypoints/pooling/
4828
+ - vllm/inputs/
4829
+ - vllm/lora/
4830
+ - vllm/model_executor/
4831
+ - vllm/multimodal/
4832
+ - vllm/outputs.py
4833
+ - vllm/platforms/
4834
+ - vllm/pooling_params.py
4835
+ - vllm/profiler/
4836
+ - vllm/sampling_params.py
4837
+ - vllm/tokenizers/
4838
+ - vllm/transformers_utils/
4839
+ - vllm/utils/
4840
+ - vllm/v1/
4841
+ commands:
4842
+ - export VLLM_WORKER_MULTIPROC_METHOD=spawn
4843
+ - bash /vllm-workspace/.buildkite/scripts/install-kv-offload.sh
4844
+ - pytest -v -s v1/kv_offload
4845
+ - pytest -v -s v1/simple_kv_offload
4846
optional: true
4854
- working_dir: "/vllm-workspace/tests"
4855
- source_file_dependencies:
4856
- - vllm/
4857
- - tests/models/multimodal/generation
4858
- commands:
4859
- - pytest -v -s models/multimodal/generation/test_common.py -m 'split(group=1) and not core_model'
4847
4861
-- label: ":amd: (MI355) Multimodal Models (Extended Pooling)" # TBD
4862
- timeout_in_minutes: 75
4848
+- label: ":amd: (MI355 DPX) V1 KV Connectors Shard %N"
4849
+ timeout_in_minutes: 60
4850
+ parallelism: 4
4851
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4852
dind: false
4865
- agent_pool: mi355_1
4866
- optional: true
4867
- working_dir: "/vllm-workspace/tests"
4853
+ agent_pool: mi355_dpx
4854
+ working_dir: /vllm-workspace/tests
4855
source_file_dependencies:
4856
- vllm/
4870
- - tests/models/multimodal/pooling
4857
+ - tests/v1/core
4858
+ - tests/v1/executor
4859
+ - tests/v1/kv_offload
4860
+ - tests/v1/simple_kv_offload
4861
+ - tests/v1/worker
4862
+ - tests/v1/streaming_input
4863
+ - tests/v1/kv_connector/unit
4864
+ - tests/v1/ec_connector/unit
4865
+ - tests/v1/metrics
4866
+ - tests/entrypoints/openai/correctness/test_lmeval.py
4867
+ - vllm/config/
4868
+ - vllm/distributed/
4869
+ - vllm/engine/
4870
+ - vllm/entrypoints/pooling/
4871
+ - vllm/inputs/
4872
+ - vllm/lora/
4873
+ - vllm/model_executor/
4874
+ - vllm/multimodal/
4875
+ - vllm/outputs.py
4876
+ - vllm/platforms/
4877
+ - vllm/pooling_params.py
4878
+ - vllm/profiler/
4879
+ - vllm/sampling_params.py
4880
+ - vllm/tokenizers/
4881
+ - vllm/transformers_utils/
4882
+ - vllm/utils/
4883
+ - vllm/v1/
4884
commands:
4872
- - pytest -v -s models/multimodal/pooling -m 'not core_model'
4885
+ - bash /vllm-workspace/.buildkite/scripts/install-kv-connectors.sh
4886
+ - export VLLM_WORKER_MULTIPROC_METHOD=spawn
4887
+ - pytest -v -s -m 'not cpu_test' v1/kv_connector/unit --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
4888
+ - pytest -v -s -m 'not cpu_test' v1/ec_connector/unit --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
4889
+ optional: true
4890
4874
-- label: ":amd: (MI355) Multimodal Models (Standard) 1: qwen2" # TBD
4891
+- label: ":amd: (MI355 DPX) V1 Metrics + LM Eval"
4892
timeout_in_minutes: 65
4893
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4894
dind: false
4878
- agent_pool: mi355_1
4879
- optional: true
4880
- working_dir: "/vllm-workspace/tests"
4895
+ agent_pool: mi355_dpx
4896
+ working_dir: /vllm-workspace/tests
4897
source_file_dependencies:
4898
- vllm/
4883
- - tests/models/multimodal
4899
+ - tests/v1/core
4900
+ - tests/v1/executor
4901
+ - tests/v1/kv_offload
4902
+ - tests/v1/simple_kv_offload
4903
+ - tests/v1/worker
4904
+ - tests/v1/streaming_input
4905
+ - tests/v1/kv_connector/unit
4906
+ - tests/v1/ec_connector/unit
4907
+ - tests/v1/metrics
4908
+ - tests/entrypoints/openai/correctness/test_lmeval.py
4909
+ - vllm/config/
4910
+ - vllm/distributed/
4911
+ - vllm/engine/
4912
+ - vllm/entrypoints/pooling/
4913
+ - vllm/inputs/
4914
+ - vllm/lora/
4915
+ - vllm/model_executor/
4916
+ - vllm/multimodal/
4917
+ - vllm/outputs.py
4918
+ - vllm/platforms/
4919
+ - vllm/pooling_params.py
4920
+ - vllm/profiler/
4921
+ - vllm/sampling_params.py
4922
+ - vllm/tokenizers/
4923
+ - vllm/transformers_utils/
4924
+ - vllm/utils/
4925
+ - vllm/v1/
4926
commands:
4885
- - pytest -v -s models/multimodal/generation/test_common.py -m core_model -k "qwen2"
4886
- - pytest -v -s models/multimodal/generation/test_ultravox.py -m core_model
4887
-
4888
-- label: ":amd: (MI355) Multimodal Models (Standard) 4: other + whisper" # TBD
4889
- timeout_in_minutes: 70
4890
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4891
- dind: false
4892
- agent_pool: mi355_1
4927
+ - export GIT_CONFIG_COUNT=1
4928
+ - export GIT_CONFIG_KEY_0=http.version
4929
+ - export GIT_CONFIG_VALUE_0=HTTP/1.1
4930
+ - export GIT_TERMINAL_PROMPT=0
4931
+ - export VLLM_WORKER_MULTIPROC_METHOD=spawn
4932
+ - pytest -v -s -m 'not cpu_test' v1/metrics
4933
+ - pip install -U git+https://github.com/vllm-project/lm-evaluation-harness.git@streaming-api
4934
+ - pytest -v -s entrypoints/openai/correctness/test_lmeval.py::test_lm_eval_accuracy_v1_engine
4935
optional: true
4894
- working_dir: "/vllm-workspace/tests"
4895
- source_file_dependencies:
4896
- - vllm/
4897
- - tests/models/multimodal
4898
- commands:
4899
- - pytest -v -s models/multimodal -m core_model --ignore models/multimodal/generation/test_common.py --ignore models/multimodal/generation/test_ultravox.py --ignore models/multimodal/generation/test_qwen2_5_vl.py --ignore models/multimodal/generation/test_qwen2_vl.py --ignore models/multimodal/generation/test_whisper.py --ignore models/multimodal/generation/test_mm_prefix_lm.py --ignore models/multimodal/generation/test_memory_leak.py --ignore models/multimodal/generation/test_vit_cudagraph.py --ignore models/multimodal/processing
4900
- - pytest -v -s models/multimodal/generation/test_vit_cudagraph.py -m core_model
4901
- - pytest models/multimodal/generation/test_memory_leak.py -m core_model
4902
- - cd .. && VLLM_WORKER_MULTIPROC_METHOD=spawn pytest -v -s tests/models/multimodal/generation/test_whisper.py -m core_model
4936
4904
-#----------------------------------------------------- mi355 · models / quantized -----------------------------------------------------#
4937
4906
-- label: ":amd: (MI355) Quantized Models" # TBD
4907
- timeout_in_minutes: 80
4938
+- label: ":amd: (MI355 DPX) V1 Sample"
4939
+ timeout_in_minutes: 70
4940
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4941
dind: false
4910
- agent_pool: mi355_1
4911
- working_dir: "/vllm-workspace/tests"
4912
- source_file_dependencies:
4913
- - vllm/model_executor/layers/quantization
4914
- - tests/models/quantization
4915
- - vllm/_aiter_ops.py
4916
- - vllm/platforms/rocm.py
4917
- - vllm/model_executor/model_loader/
4942
+ agent_pool: mi355_dpx
4943
optional: true
4919
- commands:
4920
- - unset VLLM_USE_V2_MODEL_RUNNER
4921
- - pytest -v -s models/quantization
4922
-
4923
-#------------------------------------------------------- mi355 · quantization --------------------------------------------------------#
4924
-
4925
-- label: ":amd: (MI355) Quantization Shard %N"
4926
- timeout_in_minutes: 115
4927
- parallelism: 4
4928
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4929
- dind: false
4930
- agent_pool: mi355_1
4944
working_dir: "/vllm-workspace/tests"
4945
source_file_dependencies:
4933
- - csrc/
4934
- - vllm/model_executor/layers/quantization
4935
- - tests/quantization
4936
- - vllm/_aiter_ops.py
4937
- - vllm/platforms/rocm.py
4938
- - vllm/model_executor/layers/fused_moe/oracle/unquantized.py
4939
- - vllm/model_executor/layers/fused_moe/unquantized_fused_moe_method.py
4940
- - tests/rocm/test_moe_weight_replay.py
4941
- optional: true
4946
+ - vllm/
4947
+ - tests/v1/sample
4948
+ - tests/v1/logits_processors
4949
+ - tests/v1/test_oracle.py
4950
+ - tests/v1/test_request.py
4951
+ - tests/v1/test_outputs.py
4952
+ - vllm/config/
4953
+ - vllm/distributed/
4954
+ - vllm/engine/
4955
+ - vllm/inputs/
4956
+ - vllm/logger.py
4957
+ - vllm/model_executor/
4958
+ - vllm/platforms/
4959
+ - vllm/sampling_params.py
4960
+ - vllm/transformers_utils/
4961
+ - vllm/utils/
4962
+ - vllm/v1/
4963
commands:
4943
- - unset VLLM_USE_V2_MODEL_RUNNER
4944
- - uv pip install --system torchao==0.17.0
4945
- - uv pip install --system conch-triton-kernels
4946
- - VLLM_TEST_FORCE_LOAD_FORMAT=auto pytest -v -s quantization/ --ignore quantization/test_blackwell_moe.py --ignore quantization/test_rocm_moe.py --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
4947
- - if [ "$$BUILDKITE_PARALLEL_JOB" = "0" ]; then pytest -v -s rocm/test_moe_weight_replay.py; fi
4948
-
4949
-# - label: Quantized MoE Test (B200-MI355) # TBD
4950
-# timeout_in_minutes: 180
4951
-# mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4952
-# dind: false
4953
-# agent_pool: mi355_1
4954
-# working_dir: "/vllm-workspace/"
4955
-# source_file_dependencies:
4956
-# - tests/quantization/test_gfx950_moe.py
4957
-# - vllm/model_executor/models/deepseek_v2.py
4958
-# - vllm/model_executor/models/gpt_oss.py
4959
-# - vllm/model_executor/models/llama4.py
4960
-# - vllm/model_executor/layers/fused_moe
4961
-# - vllm/model_executor/layers/quantization/compressed_tensors
4962
-# - vllm/model_executor/layers/quantization/modelopt.py
4963
-# - vllm/model_executor/layers/quantization/mxfp4.py
4964
-# - vllm/v1/attention/backends/triton_attn.py
4965
-# - vllm/v1/attention/backends/rocm_attn.py
4966
-# - vllm/v1/attention/backends/rocm_aiter_fa.py
4967
-# - vllm/v1/attention/backends/mla/
4968
-# - vllm/v1/attention/selector.py
4969
-# - vllm/model_executor/layers/layernorm.py
4970
-# - vllm/_aiter_ops.py
4971
-# - vllm/platforms/rocm.py
4972
-# - vllm/model_executor/model_loader/
4973
-# commands:
4974
-# - pytest -s -v tests/quantization/test_gfx950_moe.py
4975
-
4976
-#--------------------------------------------------------- mi355 · samplers ----------------------------------------------------------#
4964
+ - export VLLM_WORKER_MULTIPROC_METHOD=spawn
4965
+ - pytest -v -s v1/sample
4966
4978
-- label: ":amd: (MI355) Samplers Multimodal Beam Search"
4979
- timeout_in_minutes: 45
4967
+- label: ":amd: (MI355 DPX) V1 Logits + Oracle"
4968
+ timeout_in_minutes: 70
4969
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4970
dind: false
4982
- agent_pool: mi355_1
4983
- num_gpus: 1
4971
+ agent_pool: mi355_dpx
4972
optional: true
4985
- working_dir: "/vllm-workspace/tests"
4973
+ working_dir: /vllm-workspace/tests
4974
source_file_dependencies:
4987
- - vllm/model_executor/layers
4988
- - vllm/sampling_metadata.py
4989
- - vllm/v1/sample/
4990
- - vllm/entrypoints/generate/beam_search/
4991
- - tests/samplers
4992
- - tests/conftest.py
4993
- - vllm/_aiter_ops.py
4994
- - vllm/platforms/rocm.py
4975
+ - vllm/
4976
+ - tests/v1/sample
4977
+ - tests/v1/logits_processors
4978
+ - tests/v1/test_oracle.py
4979
+ - tests/v1/test_request.py
4980
+ - tests/v1/test_outputs.py
4981
+ - vllm/config/
4982
+ - vllm/distributed/
4983
+ - vllm/engine/
4984
+ - vllm/inputs/
4985
+ - vllm/logger.py
4986
+ - vllm/model_executor/
4987
+ - vllm/platforms/
4988
+ - vllm/sampling_params.py
4989
+ - vllm/transformers_utils/
4990
+ - vllm/utils/
4991
+ - vllm/v1/
4992
commands:
4996
- - pytest -v -s samplers/test_beam_search.py::test_beam_search_passes_multimodal_data
4993
+ - export VLLM_WORKER_MULTIPROC_METHOD=spawn
4994
+ - pytest -v -s v1/logits_processors
4995
+ - pytest -v -s v1/test_oracle.py
4996
+ - pytest -v -s v1/test_request.py
4997
+ - pytest -v -s v1/test_outputs.py
4998
4998
-#------------------------------------------------------------ mi355 · v1 -------------------------------------------------------------#
4999
5000
-- label: ":amd: (MI355) E2E Core Large Memory"
5000
+- label: ":amd: (MI355 DPX) E2E Core Large Memory"
5001
timeout_in_minutes: 65
5002
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
5003
dind: false
5004
- agent_pool: mi355_1
5004
+ agent_pool: mi355_dpx
5005
num_gpus: 1
5006
optional: true
5007
working_dir: "/vllm-workspace/tests"
@@ -5013,11 +5013,11 @@ steps:
5013
- pytest -v -s v1/e2e/general/test_kv_sharing_fast_prefill.py
5014
- pytest -v -s v1/e2e/general/test_mamba_prefix_cache.py::test_mamba_prefix_cache_mrv2 v1/e2e/general/test_mamba_prefix_cache.py::test_mamba_prefix_cache_mrv2_async
5015
5016
-- label: ":amd: (MI355) Extract Hidden States Integration Single GPU"
5016
+- label: ":amd: (MI355 DPX) Extract Hidden States Integration Single GPU"
5017
timeout_in_minutes: 45
5018
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
5019
dind: false
5020
- agent_pool: mi355_1
5020
+ agent_pool: mi355_dpx
5021
num_gpus: 1
5022
optional: true
5023
working_dir: "/vllm-workspace"
@@ -5032,11 +5032,11 @@ steps:
5032
- export VLLM_WORKER_MULTIPROC_METHOD=spawn
5033
- pytest -v -s -m 'not distributed' tests/v1/kv_connector/extract_hidden_states_integration
5034
5035
-- label: ":amd: (MI355) V1 Attention Shard %N" # TBD
5035
+- label: ":amd: (MI355 DPX) V1 Attention Shard %N" # TBD
5036
timeout_in_minutes: 125
5037
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
5038
dind: false
5039
- agent_pool: mi355_1
5039
+ agent_pool: mi355_dpx
5040
parallelism: 2
5041
working_dir: "/vllm-workspace/tests"
5042
source_file_dependencies:
@@ -5051,11 +5051,11 @@ steps:
5051
commands:
5052
- pytest -v -s v1/attention --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
5053
5054
-- label: ":amd: (MI355) V1 Core"
5054
+- label: ":amd: (MI355 DPX) V1 Core (MI355 suite)"
5055
timeout_in_minutes: 60
5056
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
5057
dind: false
5058
- agent_pool: mi355_1
5058
+ agent_pool: mi355_dpx
5059
optional: true
5060
working_dir: "/vllm-workspace/tests"
5061
source_file_dependencies:
@@ -5092,11 +5092,11 @@ steps:
5092
- export VLLM_WORKER_MULTIPROC_METHOD=spawn
5093
- pytest -v -s -m 'not cpu_test' v1/core
5094
5095
-- label: ":amd: (MI355) V1 Executor + Worker"
5095
+- label: ":amd: (MI355 DPX) V1 Executor + Worker (MI355 suite)"
5096
timeout_in_minutes: 60
5097
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
5098
dind: false
5099
- agent_pool: mi355_1
5099
+ agent_pool: mi355_dpx
5100
optional: true
5101
working_dir: /vllm-workspace/tests
5102
source_file_dependencies:
@@ -5135,11 +5135,11 @@ steps:
5135
- pytest -v -s v1/worker
5136
- pytest -v -s v1/streaming_input
5137
5138
-- label: ":amd: (MI355) V1 KV Offload"
5138
+- label: ":amd: (MI355 DPX) V1 KV Offload (MI355 suite)"
5139
timeout_in_minutes: 60
5140
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
5141
dind: false
5142
- agent_pool: mi355_1
5142
+ agent_pool: mi355_dpx
5143
optional: true
5144
working_dir: /vllm-workspace/tests
5145
source_file_dependencies:
@@ -5178,12 +5178,12 @@ steps:
5178
- pytest -v -s v1/kv_offload
5179
- pytest -v -s v1/simple_kv_offload
5180
5181
-- label: ":amd: (MI355) V1 KV Connectors Shard %N"
5181
+- label: ":amd: (MI355 DPX) V1 KV Connectors Shard %N (MI355 suite)"
5182
timeout_in_minutes: 60
5183
parallelism: 4
5184
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
5185
dind: false
5186
- agent_pool: mi355_1
5186
+ agent_pool: mi355_dpx
5187
optional: true
5188
working_dir: /vllm-workspace/tests
5189
source_file_dependencies:
@@ -5222,11 +5222,11 @@ steps:
5222
- pytest -v -s -m 'not cpu_test' v1/kv_connector/unit --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
5223
- pytest -v -s -m 'not cpu_test' v1/ec_connector/unit --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
5224
5225
-- label: ":amd: (MI355) V1 Metrics + LM Eval"
5225
+- label: ":amd: (MI355 DPX) V1 Metrics + LM Eval (MI355 suite)"
5226
timeout_in_minutes: 65
5227
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
5228
dind: false
5229
- agent_pool: mi355_1
5229
+ agent_pool: mi355_dpx
5230
optional: true
5231
working_dir: /vllm-workspace/tests
5232
source_file_dependencies:
@@ -5269,11 +5269,11 @@ steps:
5269
- pip install -U git+https://github.com/vllm-project/lm-evaluation-harness.git@streaming-api
5270
- pytest -v -s entrypoints/openai/correctness/test_lmeval.py::test_lm_eval_accuracy_v1_engine
5271
5272
-- label: ":amd: (MI355) CUDAGraph"
5272
+- label: ":amd: (MI355 DPX) CUDAGraph"
5273
timeout_in_minutes: 55
5274
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
5275
dind: false
5276
- agent_pool: mi355_1
5276
+ agent_pool: mi355_dpx
5277
optional: true
5278
working_dir: /vllm-workspace/tests
5279
source_file_dependencies:
@@ -5299,11 +5299,11 @@ steps:
5299
- pytest -v -s v1/cudagraph/test_encoder_cudagraph.py
5300
5301
5302
-- label: ":amd: (MI355) V1 Sample"
5302
+- label: ":amd: (MI355 DPX) V1 Sample (MI355 suite)"
5303
timeout_in_minutes: 70
5304
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
5305
dind: false
5306
- agent_pool: mi355_1
5306
+ agent_pool: mi355_dpx
5307
optional: true
5308
working_dir: "/vllm-workspace/tests"
5309
source_file_dependencies:
@@ -5328,11 +5328,11 @@ steps:
5328
- export VLLM_WORKER_MULTIPROC_METHOD=spawn
5329
- pytest -v -s v1/sample
5330
5331
-- label: ":amd: (MI355) V1 Logits + Oracle"
5331
+- label: ":amd: (MI355 DPX) V1 Logits + Oracle (MI355 suite)"
5332
timeout_in_minutes: 70
5333
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
5334
dind: false
5335
- agent_pool: mi355_1
5335
+ agent_pool: mi355_dpx
5336
optional: true
5337
working_dir: /vllm-workspace/tests
5338
source_file_dependencies:
@@ -5385,12 +5385,12 @@ steps:
5385
- export VLLM_WORKER_MULTIPROC_METHOD=spawn
5386
- pytest -v -s -m 'not slow_test' v1/spec_decode
5387
5388
-- label: ":amd: (MI355) Spec Decode AL MTP + Other Acceptance Nightly"
5388
+- label: ":amd: (MI355 DPX) Spec Decode AL MTP + Other Acceptance Nightly"
5389
timeout_in_minutes: 55
5390
parallelism: 3
5391
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
5392
dind: false
5393
- agent_pool: mi355_1
5393
+ agent_pool: mi355_dpx
5394
optional: true
5395
working_dir: "/vllm-workspace/tests"
5396
source_file_dependencies:
@@ -5420,11 +5420,11 @@ steps:
5420
5421
#----------------------------------------------------------- mi355 · misc ------------------------------------------------------------#
5422
5423
-- label: ":amd: (MI355) Regression" # TBD
5423
+- label: ":amd: (MI355 DPX) Regression" # TBD
5424
timeout_in_minutes: 25
5425
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
5426
dind: false
5427
- agent_pool: mi355_1
5427
+ agent_pool: mi355_dpx
5428
optional: true
5429
working_dir: "/vllm-workspace/tests"
5430
source_file_dependencies:
.buildkite/test_areas/basic_correctness.yaml
+2
-2
@@ -19,9 +19,9 @@ steps:
19
- pytest -v -s basic_correctness/test_cpu_offload.py
20
mirror:
21
amd:
22
- label: ":amd: (MI300) Basic Correctness"
22
+ label: ":amd: (MI355 DPX) Basic Correctness"
23
dind: false
24
- device: mi300_1
24
+ device: mi355_dpx
25
timeout_in_minutes: 60
26
depends_on:
27
- image-build-amd
.buildkite/test_areas/benchmarks.yaml
+2
-2
@@ -15,9 +15,9 @@ steps:
15
- pytest -v -s benchmarks/
16
mirror:
17
amd:
18
- label: ":amd: (MI300) Benchmarks CLI"
18
+ label: ":amd: (MI355 DPX) Benchmarks CLI"
19
dind: false
20
- device: mi300_1
20
+ device: mi355_dpx
21
timeout_in_minutes: 40
22
depends_on:
23
- image-build-amd
.buildkite/test_areas/compile.yaml
+4
-4
@@ -113,9 +113,9 @@ steps:
113
- 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)"
114
mirror:
115
amd:
116
- label: ":amd: (MI355) Fusion E2E Quick"
116
+ label: ":amd: (MI355 DPX) Fusion E2E Quick"
117
dind: false
118
- device: mi355_1
118
+ device: mi355_dpx
119
working_dir: "/vllm-workspace/"
120
timeout_in_minutes: 30
121
depends_on:
@@ -154,9 +154,9 @@ steps:
154
- pytest -v -s tests/compile/fusions_e2e/test_tp1_quant.py -k "llama-3"
155
mirror:
156
amd:
157
- label: ":amd: (MI355) Fusion E2E Config Sweep"
157
+ label: ":amd: (MI355 DPX) Fusion E2E Config Sweep"
158
dind: false
159
- device: mi355_1
159
+ device: mi355_dpx
160
working_dir: "/vllm-workspace/"
161
timeout_in_minutes: 40
162
depends_on:
.buildkite/test_areas/cuda.yaml
+2
-2
@@ -19,9 +19,9 @@ steps:
19
- pytest -v -s cuda/test_cuda_compatibility_path.py
20
mirror:
21
amd:
22
- label: ":amd: (MI355) CUDA Platform"
22
+ label: ":amd: (MI355 DPX) CUDA Platform"
23
dind: false
24
- device: mi355_1
24
+ device: mi355_dpx
25
working_dir: "/vllm-workspace/tests"
26
timeout_in_minutes: 25
27
depends_on:
.buildkite/test_areas/disaggregated.yaml
+1
-1
@@ -153,7 +153,7 @@ steps:
153
label: ":amd: (MI300) Hybrid SSM NixlConnector PD accuracy"
154
dind: false
155
device: mi300_4
156
- timeout_in_minutes: 60
156
+ timeout_in_minutes: 80
157
depends_on:
158
- image-build-amd
159
source_file_dependencies:
.buildkite/test_areas/distributed.yaml
+3
-3
@@ -35,9 +35,9 @@ steps:
35
- pytest -v -s distributed/test_sharded_rdt_trainer.py
36
mirror:
37
amd:
38
- label: ":amd: (MI355) Sharded RDT Weight Transfer"
38
+ label: ":amd: (MI355 DPX) Sharded RDT Weight Transfer"
39
dind: false
40
- device: mi355_1
40
+ device: mi355_dpx
41
num_devices: 1
42
timeout_in_minutes: 45
43
working_dir: /vllm-workspace/tests
@@ -224,7 +224,7 @@ steps:
224
label: ":amd: (MI355) Distributed Torchrun + Examples"
225
dind: false
226
device: mi355_4
227
- timeout_in_minutes: 45
227
+ timeout_in_minutes: 55
228
depends_on:
229
- image-build-amd
230
source_file_dependencies:
.buildkite/test_areas/e2e_integration.yaml
+2
-2
@@ -132,9 +132,9 @@ steps:
132
- bash .buildkite/scripts/scheduled_integration_test/deepseek_v2_lite_prefetch_offload.sh 0.25 200 8030
133
mirror:
134
amd:
135
- label: ":amd: (MI355) DeepSeek V2-Lite Prefetch Offload Accuracy"
135
+ label: ":amd: (MI355 DPX) DeepSeek V2-Lite Prefetch Offload Accuracy"
136
dind: false
137
- device: mi355_1
137
+ device: mi355_dpx
138
working_dir: "/vllm-workspace"
139
timeout_in_minutes: 30
140
depends_on:
.buildkite/test_areas/engine.yaml
+4
-4
@@ -29,9 +29,9 @@ 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"
32
+ label: ":amd: (MI355 DPX) Engine Core"
33
dind: false
34
- device: mi300_1
34
+ device: mi355_dpx
35
timeout_in_minutes: 40
36
depends_on:
37
- image-build-amd
@@ -117,9 +117,9 @@ steps:
117
- pytest -v -s v1/e2e/general/test_mamba_prefix_cache.py::test_mamba_prefix_cache_mrv2 v1/e2e/general/test_mamba_prefix_cache.py::test_mamba_prefix_cache_mrv2_async
118
mirror:
119
amd:
120
- label: ":amd: (MI355) E2E Core Large Memory"
120
+ label: ":amd: (MI355 DPX) E2E Core Large Memory"
121
dind: false
122
- device: mi355_1
122
+ device: mi355_dpx
123
num_devices: 1
124
timeout_in_minutes: 65
125
working_dir: /vllm-workspace/tests
.buildkite/test_areas/entrypoints.yaml
+22
-22
@@ -39,9 +39,9 @@ steps:
39
- pytest -v -s entrypoints/launchers
40
mirror:
41
amd:
42
- label: ":amd: (MI355) Entrypoints Unit"
42
+ label: ":amd: (MI355 DPX) Entrypoints Unit"
43
dind: false
44
- device: mi355_1
44
+ device: mi355_dpx
45
timeout_in_minutes: 50
46
depends_on:
47
- image-build-amd
@@ -64,9 +64,9 @@ steps:
64
- pytest -v -s entrypoints/llm/offline_mode # Needs to avoid interference with other tests
65
mirror:
66
amd:
67
- label: ":amd: (MI300) Entrypoints Integration (LLM)"
67
+ label: ":amd: (MI355 DPX) Entrypoints Integration (LLM)"
68
dind: false
69
- device: mi300_1
69
+ device: mi355_dpx
70
timeout_in_minutes: 90
71
depends_on:
72
- image-build-amd
@@ -89,9 +89,9 @@ steps:
89
- pytest -v -s entrypoints/scale_out --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
90
mirror:
91
amd:
92
- label: ":amd: (MI300) Entrypoints Integration (API Server) %N"
92
+ label: ":amd: (MI355 DPX) Entrypoints Integration (API Server) %N"
93
dind: false
94
- device: mi300_1
94
+ device: mi355_dpx
95
timeout_in_minutes: 65
96
depends_on:
97
- image-build-amd
@@ -111,9 +111,9 @@ steps:
111
- pytest -v -s entrypoints/openai --ignore=entrypoints/openai/completion --ignore=entrypoints/openai/chat_completion --ignore=entrypoints/openai/responses --ignore=entrypoints/openai/correctness
112
mirror:
113
amd:
114
- label: ":amd: (MI300) Entrypoints Integration (OpenAI API completion)"
114
+ label: ":amd: (MI355 DPX) Entrypoints Integration (OpenAI API completion)"
115
dind: false
116
- device: mi300_1
116
+ device: mi355_dpx
117
timeout_in_minutes: 65
118
depends_on:
119
- image-build-amd
@@ -132,9 +132,9 @@ steps:
132
- pytest -v -s entrypoints/openai/chat_completion
133
mirror:
134
amd:
135
- label: ":amd: (MI300) Entrypoints Integration (OpenAI API chat_completion)"
135
+ label: ":amd: (MI355 DPX) Entrypoints Integration (OpenAI API chat_completion)"
136
dind: false
137
- device: mi300_1
137
+ device: mi355_dpx
138
timeout_in_minutes: 70
139
depends_on:
140
- image-build-amd
@@ -160,9 +160,9 @@ steps:
160
- pytest -v -s entrypoints/cohere
161
mirror:
162
amd:
163
- label: ":amd: (MI300) Entrypoints Integration (API Server Generate)"
163
+ label: ":amd: (MI355 DPX) Entrypoints Integration (API Server Generate)"
164
dind: false
165
- device: mi300_1
165
+ device: mi355_dpx
166
timeout_in_minutes: 65
167
depends_on:
168
- image-build-amd
@@ -180,9 +180,9 @@ steps:
180
- pytest -v -s entrypoints/openai/responses
181
mirror:
182
amd:
183
- label: ":amd: (MI355) Entrypoints Integration (Responses API)"
183
+ label: ":amd: (MI355 DPX) Entrypoints Integration (Responses API)"
184
dind: false
185
- device: mi355_1
185
+ device: mi355_dpx
186
timeout_in_minutes: 50
187
depends_on:
188
- image-build-amd
@@ -201,9 +201,9 @@ steps:
201
- pytest -v -s entrypoints/speech_to_text
202
mirror:
203
amd:
204
- label: ":amd: (MI355) Entrypoints Integration (Speech to Text)"
204
+ label: ":amd: (MI355 DPX) Entrypoints Integration (Speech to Text)"
205
dind: false
206
- device: mi355_1
206
+ device: mi355_dpx
207
timeout_in_minutes: 60
208
working_dir: "/vllm-workspace/tests"
209
depends_on:
@@ -223,9 +223,9 @@ steps:
223
- pytest -v -s entrypoints/multimodal
224
mirror:
225
amd:
226
- label: ":amd: (MI355) Entrypoints Integration (Multimodal)"
226
+ label: ":amd: (MI355 DPX) Entrypoints Integration (Multimodal)"
227
dind: false
228
- device: mi355_1
228
+ device: mi355_dpx
229
timeout_in_minutes: 55
230
depends_on:
231
- image-build-amd
@@ -244,9 +244,9 @@ steps:
244
- pytest -v -s entrypoints/pooling
245
mirror:
246
amd:
247
- label: ":amd: (MI355) Entrypoints Integration (Pooling)"
247
+ label: ":amd: (MI355 DPX) Entrypoints Integration (Pooling)"
248
dind: false
249
- device: mi355_1
249
+ device: mi355_dpx
250
timeout_in_minutes: 65
251
working_dir: "/vllm-workspace/tests"
252
depends_on:
@@ -263,9 +263,9 @@ steps:
263
- pytest -s entrypoints/openai/correctness/
264
mirror:
265
amd:
266
- label: ":amd: (MI300) OpenAI API Correctness"
266
+ label: ":amd: (MI355 DPX) OpenAI API Correctness"
267
dind: false
268
- device: mi300_1
268
+ device: mi355_dpx
269
timeout_in_minutes: 30
270
depends_on:
271
- image-build-amd
.buildkite/test_areas/expert_parallelism.yaml
+2
-2
@@ -16,9 +16,9 @@ steps:
16
- pytest -v -s distributed/test_eplb_utils.py
17
mirror:
18
amd:
19
- label: ":amd: (MI300) EPLB Algorithm"
19
+ label: ":amd: (MI355 DPX) EPLB Algorithm"
20
dind: false
21
- device: mi300_1
21
+ device: mi355_dpx
22
timeout_in_minutes: 30
23
depends_on:
24
- image-build-amd
.buildkite/test_areas/kernels.yaml
+70
-14
@@ -15,9 +15,9 @@ steps:
15
- pytest -v -s tests/kernels/ir
16
mirror:
17
amd:
18
- label: ":amd: (MI355) vLLM IR"
18
+ label: ":amd: (MI355 DPX) vLLM IR"
19
dind: false
20
- device: mi355_1
20
+ device: mi355_dpx
21
timeout_in_minutes: 50
22
working_dir: "/vllm-workspace/"
23
depends_on:
@@ -44,9 +44,9 @@ steps:
44
parallelism: 3
45
mirror:
46
amd:
47
- label: ":amd: (MI355) Core Operation Kernels Shard %N"
47
+ label: ":amd: (MI300) Core Operation Kernels Shard %N"
48
dind: false
49
- device: mi355_1
49
+ device: mi300_1
50
timeout_in_minutes: 40
51
working_dir: "/vllm-workspace/tests"
52
depends_on:
@@ -113,9 +113,9 @@ steps:
113
- pytest -v -s kernels/test_top_k_per_row.py
114
mirror:
115
amd:
116
- label: ":amd: (MI355) DeepSeek V4 Fused Kernels"
116
+ label: ":amd: (MI355 DPX) DeepSeek V4 Fused Kernels"
117
dind: false
118
- device: mi355_1
118
+ device: mi355_dpx
119
num_devices: 1
120
timeout_in_minutes: 60
121
working_dir: /vllm-workspace/tests
@@ -230,9 +230,9 @@ steps:
230
- pytest -v -s kernels/attention/test_triton_unified_attention_diffkv.py
231
mirror:
232
amd:
233
- label: ":amd: (MI355) Attention DiffKV Kernels"
233
+ label: ":amd: (MI355 DPX) Attention DiffKV Kernels"
234
dind: false
235
- device: mi355_1
235
+ device: mi355_dpx
236
num_devices: 1
237
timeout_in_minutes: 35
238
working_dir: /vllm-workspace/tests
@@ -270,9 +270,9 @@ steps:
270
- pytest -v -s kernels/attention/test_mla_cross_layer_kernel_equivalence.py
271
mirror:
272
amd:
273
- label: ":amd: (MI355) MLA Kernels"
273
+ label: ":amd: (MI355 DPX) MLA Kernels"
274
dind: false
275
- device: mi355_1
275
+ device: mi355_dpx
276
num_devices: 1
277
timeout_in_minutes: 40
278
working_dir: /vllm-workspace/
@@ -768,10 +768,66 @@ steps:
768
- amd-smi
769
- VLLM_ROCM_USE_AITER=1 VLLM_ROCM_USE_AITER_MOE=1 VLLM_ROCM_USE_AITER_FUSION_SHARED_EXPERTS=0 pytest -v -s kernels/moe/test_mori_moe.py
770
771
-- label: ":amd: (MI355) FP8 MoE Kernels"
771
+- label: ":amd: (MI355 DPX) Native Quantization Kernels"
772
+ key: amd-native-quantization-kernels-mi355
773
+ dind: false
774
+ device: mi355_dpx
775
+ num_devices: 1
776
+ timeout_in_minutes: 40
777
+ working_dir: /vllm-workspace/tests
778
+ depends_on:
779
+ - image-build-amd
780
+ source_file_dependencies:
781
+ - csrc/rocm/
782
+ - vllm/model_executor/layers/quantization/
783
+ - vllm/model_executor/kernels/linear/
784
+ - vllm/_aiter_ops.py
785
+ - vllm/_custom_ops.py
786
+ - vllm/platforms/rocm.py
787
+ - vllm/envs.py
788
+ - tests/kernels/quantization/test_rocm_mxfp4.py
789
+ - tests/kernels/quantization/test_rocm_mxfp8_linear.py
790
+ - tests/kernels/quantization/test_rocm_skinny_gemms.py
791
+ - tests/quantization/reference_mxfp4.py
792
+ - tests/kernels/quant_utils.py
793
+ - tests/kernels/utils.py
794
+ commands:
795
+ - pytest -v -s kernels/quantization/test_rocm_mxfp4.py
796
+ - pytest -v -s kernels/quantization/test_rocm_mxfp8_linear.py
797
+ - pytest -v -s kernels/quantization/test_rocm_skinny_gemms.py -k wvsplitkrc
798
+
799
+- label: ":amd: (MI355 DPX) FP4 and INT4 MoE Kernels"
800
+ key: amd-fp4-moe-kernels-mi355
801
+ dind: false
802
+ device: mi355_dpx
803
+ num_devices: 1
804
+ timeout_in_minutes: 40
805
+ working_dir: /vllm-workspace/tests
806
+ depends_on:
807
+ - image-build-amd
808
+ source_file_dependencies:
809
+ - csrc/moe/
810
+ - csrc/rocm/
811
+ - vllm/model_executor/layers/fused_moe/
812
+ - vllm/model_executor/layers/quantization/
813
+ - vllm/_aiter_ops.py
814
+ - vllm/platforms/rocm.py
815
+ - vllm/envs.py
816
+ - vllm/config/
817
+ - tests/kernels/moe/test_rocm_aiter_moe.py
818
+ - tests/kernels/moe/test_ocp_mx_moe.py
819
+ - tests/kernels/moe/test_flydsl_moe.py
820
+ - tests/kernels/moe/utils.py
821
+ - tests/kernels/utils.py
822
+ commands:
823
+ - pytest -v -s kernels/moe/test_rocm_aiter_moe.py -k 'mi350_mxfp4 or quant_scheme_support_matches_gfx950'
824
+ - pytest -v -s kernels/moe/test_ocp_mx_moe.py::test_rocm_mxfp4_moe_oracle -k AITER_MXFP4
825
+ - pytest -v -s kernels/moe/test_flydsl_moe.py
826
+
827
+- label: ":amd: (MI355 DPX) FP8 MoE Kernels"
828
key: amd-fp8-moe-kernels-mi355
829
dind: false
774
- device: mi355_1
830
+ device: mi355_dpx
831
num_devices: 1
832
timeout_in_minutes: 40
833
working_dir: /vllm-workspace/tests
@@ -798,10 +854,10 @@ steps:
854
- pytest -v -s kernels/moe/test_triton_moe_no_act_mul.py
855
- pytest -v -s kernels/moe/test_triton_moe_ptpc_fp8.py
856
801
-- label: ":amd: (MI355) Kernels"
857
+- label: ":amd: (MI355 DPX) Kernels"
858
key: amd-kernels-mi355
859
dind: false
804
- device: mi355_1
860
+ device: mi355_dpx
861
num_devices: 1
862
timeout_in_minutes: 25
863
working_dir: /vllm-workspace/
.buildkite/test_areas/lm_eval.yaml
+26
-4
@@ -68,9 +68,9 @@ steps:
68
- pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-watermark.txt
69
mirror:
70
amd:
71
- label: ":amd: (MI355) LM Eval Watermarking"
71
+ label: ":amd: (MI355 DPX) LM Eval Watermarking"
72
dind: false
73
- device: mi355_1
73
+ device: mi355_dpx
74
num_devices: 1
75
timeout_in_minutes: 45
76
working_dir: /vllm-workspace/tests
@@ -146,6 +146,28 @@ steps:
146
- vllm/model_executor/layers/quantization
147
commands:
148
- pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-blackwell.txt
149
+ mirror:
150
+ amd:
151
+ label: ":amd: (MI355 DPX) LM Eval Small Models"
152
+ dind: false
153
+ device: mi355_dpx
154
+ timeout_in_minutes: 45
155
+ working_dir: /vllm-workspace/tests
156
+ depends_on:
157
+ - image-build-amd
158
+ source_file_dependencies:
159
+ - csrc/
160
+ - tests/evals/gsm8k/
161
+ - vllm/model_executor/layers/quantization
162
+ - vllm/model_executor/kernels/linear/
163
+ - vllm/model_executor/models/
164
+ - vllm/model_executor/model_loader/
165
+ - vllm/v1/attention/backends/
166
+ - vllm/v1/attention/selector.py
167
+ - vllm/_aiter_ops.py
168
+ - vllm/platforms/rocm.py
169
+ commands:
170
+ - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-gfx950-small.txt
171
172
- label: ":nvidia: (B200) LM Eval Small Models Distributed"
173
key: lm-eval-small-models-distributed-2xb200
@@ -972,9 +994,9 @@ steps:
994
- pytest -s -v evals/mrcr/test_mrcr_correctness.py --config-list-file=evals/mrcr/configs/models-small.txt
995
mirror:
996
amd:
975
- label: ":amd: (MI355) MRCR Eval Small Models"
997
+ label: ":amd: (MI355 DPX) MRCR Eval Small Models"
998
dind: false
977
- device: mi355_1
999
+ device: mi355_dpx
1000
timeout_in_minutes: 40
1001
depends_on:
1002
- image-build-amd
.buildkite/test_areas/lora.yaml
+2
-2
@@ -14,9 +14,9 @@ steps:
14
parallelism: 4
15
mirror:
16
amd:
17
- label: ":amd: (MI300) LoRA Shard %N"
17
+ label: ":amd: (MI355 DPX) LoRA Shard %N"
18
dind: false
19
- device: mi300_1
19
+ device: mi355_dpx
20
working_dir: "/vllm-workspace/tests"
21
timeout_in_minutes: 85
22
source_file_dependencies:
.buildkite/test_areas/misc.yaml
+24
-24
@@ -29,9 +29,9 @@ steps:
29
- pytest -v -s v1/sample
30
mirror:
31
amd:
32
- label: ":amd: (MI300) V1 Sample"
32
+ label: ":amd: (MI355 DPX) V1 Sample"
33
dind: false
34
- device: mi300_1
34
+ device: mi355_dpx
35
timeout_in_minutes: 70
36
depends_on:
37
- image-build-amd
@@ -51,9 +51,9 @@ steps:
51
- pytest -v -s v1/test_outputs.py
52
mirror:
53
amd:
54
- label: ":amd: (MI300) V1 Logits + Oracle"
54
+ label: ":amd: (MI355 DPX) V1 Logits + Oracle"
55
dind: false
56
- device: mi300_1
56
+ device: mi355_dpx
57
timeout_in_minutes: 70
58
depends_on:
59
- image-build-amd
@@ -96,9 +96,9 @@ steps:
96
- pytest -v -s -m 'not cpu_test' v1/core
97
mirror:
98
amd:
99
- label: ":amd: (MI300) V1 Core"
99
+ label: ":amd: (MI355 DPX) V1 Core"
100
dind: false
101
- device: mi300_1
101
+ device: mi355_dpx
102
timeout_in_minutes: 60
103
depends_on:
104
- image-build-amd
@@ -116,9 +116,9 @@ steps:
116
- pytest -v -s v1/streaming_input
117
mirror:
118
amd:
119
- label: ":amd: (MI300) V1 Executor + Worker"
119
+ label: ":amd: (MI355 DPX) V1 Executor + Worker"
120
dind: false
121
- device: mi300_1
121
+ device: mi355_dpx
122
timeout_in_minutes: 60
123
depends_on:
124
- image-build-amd
@@ -135,9 +135,9 @@ steps:
135
- pytest -v -s v1/simple_kv_offload
136
mirror:
137
amd:
138
- label: ":amd: (MI300) V1 KV Offload"
138
+ label: ":amd: (MI355 DPX) V1 KV Offload"
139
dind: false
140
- device: mi300_1
140
+ device: mi355_dpx
141
timeout_in_minutes: 60
142
depends_on:
143
- image-build-amd
@@ -155,9 +155,9 @@ steps:
155
- pytest -v -s -m 'not cpu_test' v1/ec_connector/unit --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
156
mirror:
157
amd:
158
- label: ":amd: (MI300) V1 KV Connectors Shard %N"
158
+ label: ":amd: (MI355 DPX) V1 KV Connectors Shard %N"
159
dind: false
160
- device: mi300_1
160
+ device: mi355_dpx
161
timeout_in_minutes: 60
162
depends_on:
163
- image-build-amd
@@ -182,9 +182,9 @@ steps:
182
- pytest -v -s entrypoints/openai/correctness/test_lmeval.py::test_lm_eval_accuracy_v1_engine
183
mirror:
184
amd:
185
- label: ":amd: (MI300) V1 Metrics + LM Eval"
185
+ label: ":amd: (MI355 DPX) V1 Metrics + LM Eval"
186
dind: false
187
- device: mi300_1
187
+ device: mi355_dpx
188
timeout_in_minutes: 65
189
depends_on:
190
- image-build-amd
@@ -239,9 +239,9 @@ steps:
239
- pytest -v -s tests/v1/kv_connector/extract_hidden_states_integration
240
mirror:
241
amd:
242
- label: ":amd: (MI355) Extract Hidden States Integration Single GPU"
242
+ label: ":amd: (MI355 DPX) Extract Hidden States Integration Single GPU"
243
dind: false
244
- device: mi355_1
244
+ device: mi355_dpx
245
num_devices: 1
246
timeout_in_minutes: 45
247
working_dir: /vllm-workspace
@@ -328,9 +328,9 @@ steps:
328
working_dir: "/vllm-workspace/tests" # optional
329
mirror:
330
amd:
331
- label: ":amd: (MI355) Regression"
331
+ label: ":amd: (MI355 DPX) Regression"
332
dind: false
333
- device: mi355_1
333
+ device: mi355_dpx
334
timeout_in_minutes: 25
335
depends_on:
336
- image-build-amd
@@ -370,9 +370,9 @@ steps:
370
- 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
371
mirror:
372
amd:
373
- label: ":amd: (MI300) Examples"
373
+ label: ":amd: (MI355 DPX) Examples"
374
dind: false
375
- device: mi300_1
375
+ device: mi355_dpx
376
timeout_in_minutes: 75
377
source_file_dependencies:
378
- vllm/entrypoints
@@ -472,9 +472,9 @@ steps:
472
- pytest -v -s utils_
473
mirror:
474
amd:
475
- label: ":amd: (MI355) Async Engine, Inputs, Utils, Worker"
475
+ label: ":amd: (MI355 DPX) Async Engine, Inputs, Utils, Worker"
476
dind: false
477
- device: mi355_1
477
+ device: mi355_dpx
478
timeout_in_minutes: 50
479
depends_on:
480
- image-build-amd
@@ -667,9 +667,9 @@ steps:
667
- pytest -v -s v1/spec_decode/test_acceptance_length.py -m slow_test
668
mirror:
669
amd:
670
- label: ":amd: (MI355) Acceptance Length (Large Models)"
670
+ label: ":amd: (MI355 DPX) Acceptance Length (Large Models)"
671
dind: false
672
- device: mi355_1
672
+ device: mi355_dpx
673
timeout_in_minutes: 45
674
depends_on:
675
- image-build-amd
.buildkite/test_areas/models_basic.yaml
+10
-10
@@ -16,9 +16,9 @@ steps:
16
- pytest -v -s models/test_initialization.py::test_can_initialize_small_subset
17
mirror:
18
amd:
19
- label: ":amd: (MI355) Basic Models (Initialization)"
19
+ label: ":amd: (MI355 DPX) Basic Models (Initialization)"
20
dind: false
21
- device: mi355_1
21
+ device: mi355_dpx
22
timeout_in_minutes: 30
23
depends_on:
24
- image-build-amd
@@ -39,9 +39,9 @@ steps:
39
parallelism: 14
40
mirror:
41
amd:
42
- label: ":amd: (MI355) Basic Models (Extra Initialization) Shard %N"
42
+ label: ":amd: (MI355 DPX) Basic Models (Extra Initialization) Shard %N"
43
dind: false
44
- device: mi355_1
44
+ device: mi355_dpx
45
timeout_in_minutes: 60
46
parallelism: 10
47
working_dir: "/vllm-workspace/tests"
@@ -70,9 +70,9 @@ steps:
70
- pytest -v -s models/test_terratorch.py models/transformers/test_backend.py models/test_registry.py models/test_deepseek_v4_vl_rocm.py
71
mirror:
72
amd:
73
- label: ":amd: (MI300) Basic Models (Other)"
73
+ label: ":amd: (MI355 DPX) Basic Models (Other)"
74
dind: false
75
- device: mi300_1
75
+ device: mi355_dpx
76
timeout_in_minutes: 65
77
depends_on:
78
- image-build-amd
@@ -138,9 +138,9 @@ steps:
138
- pytest -v -s models/glm5next
139
mirror:
140
amd:
141
- label: ":amd: (MI355) GLM5Next"
141
+ label: ":amd: (MI355 DPX) GLM5Next"
142
dind: false
143
- device: mi355_1
143
+ device: mi355_dpx
144
timeout_in_minutes: 35
145
working_dir: "/vllm-workspace/tests"
146
depends_on:
@@ -162,9 +162,9 @@ steps:
162
- pytest -v -s models/qwen4_exp/test_hc_ops.py models/qwen4_exp/test_qsa_pre_indexer.py models/qwen4_exp/test_qsa_reference.py
163
mirror:
164
amd:
165
- label: ":amd: (MI300) Qwen4 Exp"
165
+ label: ":amd: (MI355 DPX) Qwen4 Exp"
166
dind: false
167
- device: mi300_1
167
+ device: mi355_dpx
168
depends_on:
169
- image-build-amd
170
commands:
.buildkite/test_areas/models_language.yaml
+10
-10
@@ -16,9 +16,9 @@ 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)"
19
+ label: ":amd: (MI355 DPX) Language Models (Standard)"
20
dind: false
21
- device: mi300_1
21
+ device: mi355_dpx
22
timeout_in_minutes: 60
23
depends_on:
24
- image-build-amd
@@ -40,9 +40,9 @@ steps:
40
parallelism: 2
41
mirror:
42
amd:
43
- label: ":amd: (MI300) Language Models (Extra Standard) Shard %N"
43
+ label: ":amd: (MI355 DPX) Language Models (Extra Standard) Shard %N"
44
dind: false
45
- device: mi300_1
45
+ device: mi355_dpx
46
timeout_in_minutes: 40
47
depends_on:
48
- image-build-amd
@@ -82,9 +82,9 @@ steps:
82
parallelism: 2
83
mirror:
84
amd:
85
- label: ":amd: (MI355) Language Models (Hybrid) Shard %N"
85
+ label: ":amd: (MI355 DPX) Language Models (Hybrid) Shard %N"
86
dind: false
87
- device: mi355_1
87
+ device: mi355_dpx
88
timeout_in_minutes: 65
89
depends_on:
90
- image-build-amd
@@ -113,9 +113,9 @@ steps:
113
- pytest -v -s models/language/generation -m hybrid_model -k 'granite-4.0-tiny-preview'
114
mirror:
115
amd:
116
- label: ":amd: (MI355) Granite Language Model Compatibility"
116
+ label: ":amd: (MI355 DPX) Granite Language Model Compatibility"
117
dind: false
118
- device: mi355_1
118
+ device: mi355_dpx
119
num_devices: 1
120
timeout_in_minutes: 80
121
working_dir: /vllm-workspace/tests
@@ -151,9 +151,9 @@ steps:
151
- pytest -v -s models/language/generation -m '(not core_model) and (not hybrid_model)'
152
mirror:
153
amd:
154
- label: ":amd: (MI355) Language Models (Extended Generation)"
154
+ label: ":amd: (MI355 DPX) Language Models (Extended Generation)"
155
dind: false
156
- device: mi355_1
156
+ device: mi355_dpx
157
timeout_in_minutes: 95
158
depends_on:
159
- image-build-amd
.buildkite/test_areas/models_multimodal.yaml
+14
-14
@@ -15,9 +15,9 @@ 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"
18
+ label: ":amd: (MI355 DPX) Multimodal Models (Standard) 1: qwen2"
19
dind: false
20
- device: mi300_1
20
+ device: mi355_dpx
21
timeout_in_minutes: 65
22
depends_on:
23
- image-build-amd
@@ -36,9 +36,9 @@ 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"
39
+ label: ":amd: (MI355 DPX) Multimodal Models (Standard) 2: qwen3 + gemma"
40
dind: false
41
- device: mi300_1
41
+ device: mi355_dpx
42
timeout_in_minutes: 55
43
depends_on:
44
- image-build-amd
@@ -78,9 +78,9 @@ steps:
78
- 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
79
mirror:
80
amd:
81
- label: ":amd: (MI355) Multimodal Models (Standard) 4: other + whisper"
81
+ label: ":amd: (MI355 DPX) Multimodal Models (Standard) 4: other + whisper"
82
dind: false
83
- device: mi355_1
83
+ device: mi355_dpx
84
timeout_in_minutes: 70
85
depends_on:
86
- image-build-amd
@@ -125,9 +125,9 @@ steps:
125
- pytest -v -s models/multimodal/processing/test_tensor_schema.py --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
126
mirror:
127
amd:
128
- label: ":amd: (MI355) Multimodal Processor Shard %N"
128
+ label: ":amd: (MI355 DPX) Multimodal Processor Shard %N"
129
dind: false
130
- device: mi355_1
130
+ device: mi355_dpx
131
parallelism: 4
132
timeout_in_minutes: 115
133
depends_on:
@@ -146,9 +146,9 @@ steps:
146
- pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-mm-small.txt --tp-size=1
147
mirror:
148
amd:
149
- label: ":amd: (MI300) Multimodal Accuracy Eval (Small Models)"
149
+ label: ":amd: (MI355 DPX) Multimodal Accuracy Eval (Small Models)"
150
dind: false
151
- device: mi300_1
151
+ device: mi355_dpx
152
timeout_in_minutes: 35
153
depends_on:
154
- image-build-amd
@@ -212,9 +212,9 @@ steps:
212
- 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
213
mirror:
214
amd:
215
- label: ":amd: (MI355) Multimodal Models (Extended Generation 2) Shard %N"
215
+ label: ":amd: (MI355 DPX) Multimodal Models (Extended Generation 2) Shard %N"
216
dind: false
217
- device: mi355_1
217
+ device: mi355_dpx
218
timeout_in_minutes: 100
219
depends_on:
220
- image-build-amd
@@ -231,9 +231,9 @@ steps:
231
- pytest -v -s models/multimodal/generation/test_common.py -m 'split(group=1) and not core_model'
232
mirror:
233
amd:
234
- label: ":amd: (MI355) Multimodal Models (Extended Generation 3)"
234
+ label: ":amd: (MI355 DPX) Multimodal Models (Extended Generation 3)"
235
dind: false
236
- device: mi355_1
236
+ device: mi355_dpx
237
timeout_in_minutes: 90
238
depends_on:
239
- image-build-amd
.buildkite/test_areas/plugins.yaml
+2
-2
@@ -78,9 +78,9 @@ steps:
78
- pytest -v -s plugins_tests/gguf
79
mirror:
80
amd:
81
- label: ":amd: (MI355) GGUF Plugin"
81
+ label: ":amd: (MI355 DPX) GGUF Plugin"
82
dind: false
83
- device: mi355_1
83
+ device: mi355_dpx
84
timeout_in_minutes: 45
85
working_dir: "/vllm-workspace/tests"
86
depends_on:
.buildkite/test_areas/pytorch.yaml
+8
-4
@@ -89,9 +89,9 @@ steps:
89
- "find compile/h100/ -name 'test_*.py' -print0 | xargs -0 -n1 -I{} pytest -s -v '{}'"
90
mirror:
91
amd:
92
- label: ":amd: (MI355) PyTorch Compilation H100 Cases"
92
+ label: ":amd: (MI355 DPX) PyTorch Compilation H100 Cases"
93
dind: false
94
- device: mi355_1
94
+ device: mi355_dpx
95
timeout_in_minutes: 45
96
working_dir: "/vllm-workspace/tests"
97
depends_on:
@@ -192,10 +192,14 @@ steps:
192
- "find compile/fullgraph/ -name 'test_*.py' -not -name 'test_full_cudagraph.py' -print0 | xargs -0 -n1 -I{} pytest -s -v '{}'"
193
mirror:
194
amd:
195
- label: ":amd: (MI355) PyTorch Fullgraph"
195
+ label: ":amd: (MI355 DPX) PyTorch Fullgraph"
196
dind: false
197
- device: mi355_1
197
+ device: mi355_dpx
198
timeout_in_minutes: 90
199
+ # Remove once the ROCm Triton pin includes triton-lang/triton#11567.
200
+ # Avoid overlapping LLVM codegen/LLD by disabling background precompilation.
201
+ env:
202
+ TORCHINDUCTOR_PRECOMPILATION_TIMEOUT_SECONDS: "0"
203
depends_on:
204
- image-build-amd
205
.buildkite/test_areas/quantization.yaml
+4
-4
@@ -22,9 +22,9 @@ steps:
22
- VLLM_TEST_FORCE_LOAD_FORMAT=auto pytest -v -s quantization/ --ignore quantization/test_blackwell_moe.py --ignore quantization/test_rocm_moe.py -k 'not test_compressed_tensors_w4a8_fp8' --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
23
mirror:
24
amd:
25
- label: ":amd: (MI355) Quantization Shard %N"
25
+ label: ":amd: (MI300) Quantization Shard %N"
26
dind: false
27
- device: mi355_1
27
+ device: mi300_1
28
timeout_in_minutes: 115
29
working_dir: "/vllm-workspace/tests"
30
depends_on:
@@ -116,9 +116,9 @@ steps:
116
- pytest -v -s models/quantization
117
mirror:
118
amd:
119
- label: ":amd: (MI355) Quantized Models"
119
+ label: ":amd: (MI355 DPX) Quantized Models"
120
dind: false
121
- device: mi355_1
121
+ device: mi355_dpx
122
timeout_in_minutes: 80
123
working_dir: "/vllm-workspace/tests"
124
depends_on:
.buildkite/test_areas/ray_compat.yaml
+2
-2
@@ -18,9 +18,9 @@ steps:
18
- bash /vllm-workspace/.buildkite/scripts/check-ray-compatibility.sh
19
mirror:
20
amd:
21
- label: ":amd: (MI355) Ray Dependency Compatibility Check"
21
+ label: ":amd: (MI355 DPX) Ray Dependency Compatibility Check"
22
dind: false
23
- device: mi355_1
23
+ device: mi355_dpx
24
timeout_in_minutes: 25
25
working_dir: "/"
26
depends_on:
.buildkite/test_areas/rust_frontend.yaml
+8
-8
@@ -43,9 +43,9 @@ steps:
43
- pytest -v -s v1/sample/test_logprobs_e2e.py -k "test_prompt_logprobs_e2e_server"
44
mirror:
45
amd:
46
- label: ":amd: (MI355) Rust Frontend OpenAI Coverage"
46
+ label: ":amd: (MI355 DPX) Rust Frontend OpenAI Coverage"
47
dind: false
48
- device: mi355_1
48
+ device: mi355_dpx
49
timeout_in_minutes: 35
50
depends_on:
51
- image-build-amd
@@ -98,9 +98,9 @@ steps:
98
- pytest -v -s entrypoints/serve/tokenize/test_tokenization.py -k "not tokenizer_info"
99
mirror:
100
amd:
101
- label: ":amd: (MI355) Rust Frontend Serve/Admin Coverage"
101
+ label: ":amd: (MI355 DPX) Rust Frontend Serve/Admin Coverage"
102
dind: false
103
- device: mi355_1
103
+ device: mi355_dpx
104
timeout_in_minutes: 30
105
depends_on:
106
- image-build-amd
@@ -133,9 +133,9 @@ steps:
133
- pytest -s entrypoints/openai/correctness/test_lmeval.py::test_lm_eval_accuracy_v1_engine
134
mirror:
135
amd:
136
- label: ":amd: (MI355) Rust Frontend Core Correctness"
136
+ label: ":amd: (MI355 DPX) Rust Frontend Core Correctness"
137
dind: false
138
- device: mi355_1
138
+ device: mi355_dpx
139
timeout_in_minutes: 25
140
depends_on:
141
- image-build-amd
@@ -163,9 +163,9 @@ steps:
163
- 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"
164
mirror:
165
amd:
166
- label: ":amd: (MI355) Rust Frontend Tool Use"
166
+ label: ":amd: (MI355 DPX) Rust Frontend Tool Use"
167
dind: false
168
- device: mi355_1
168
+ device: mi355_dpx
169
timeout_in_minutes: 30
170
depends_on:
171
- image-build-amd
.buildkite/test_areas/samplers.yaml
+2
-2
@@ -53,9 +53,9 @@ steps:
53
- VLLM_USE_FLASHINFER_SAMPLER=1 pytest -v -s samplers/test_beam_search.py::test_beam_search_passes_multimodal_data
54
mirror:
55
amd:
56
- label: ":amd: (MI355) Samplers Multimodal Beam Search"
56
+ label: ":amd: (MI355 DPX) Samplers Multimodal Beam Search"
57
dind: false
58
- device: mi355_1
58
+ device: mi355_dpx
59
num_devices: 1
60
timeout_in_minutes: 45
61
working_dir: /vllm-workspace/tests
.buildkite/test_areas/spec_decode.yaml
+10
-10
@@ -44,9 +44,9 @@ steps:
44
- pytest -v -s v1/e2e/spec_decode/eagle/ -k "deepseek_eagle or qwen3_eagle3"
45
mirror:
46
amd:
47
- label: ":amd: (MI300) Spec Decode Eagle 1: DeepSeek + Qwen"
47
+ label: ":amd: (MI355 DPX) Spec Decode Eagle 1: DeepSeek + Qwen"
48
dind: false
49
- device: mi300_1
49
+ device: mi355_dpx
50
timeout_in_minutes: 60
51
depends_on:
52
- image-build-amd
@@ -71,9 +71,9 @@ steps:
71
- pytest -v -s v1/e2e/spec_decode/eagle/ -k "not deepseek_eagle and not qwen3_eagle3"
72
mirror:
73
amd:
74
- label: ":amd: (MI300) Spec Decode Eagle 2: Llama 3 + Qwen VL + Other"
74
+ label: ":amd: (MI355 DPX) Spec Decode Eagle 2: Llama 3 + Qwen VL + Other"
75
dind: false
76
- device: mi300_1
76
+ device: mi355_dpx
77
timeout_in_minutes: 60
78
depends_on:
79
- image-build-amd
@@ -157,9 +157,9 @@ steps:
157
- python3 spec_decode/test_custom_proposer.py
158
mirror:
159
amd:
160
- label: ":amd: (MI300) Spec Decode N-Gram + Suffix"
160
+ label: ":amd: (MI355 DPX) Spec Decode N-Gram + Suffix"
161
dind: false
162
- device: mi300_1
162
+ device: mi355_dpx
163
timeout_in_minutes: 35
164
depends_on:
165
- image-build-amd
@@ -185,9 +185,9 @@ steps:
185
- pytest -v -s v1/e2e/spec_decode/draft_model/ --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
186
mirror:
187
amd:
188
- label: ":amd: (MI300) Spec Decode Draft Model Shard %N"
188
+ label: ":amd: (MI355 DPX) Spec Decode Draft Model Shard %N"
189
dind: false
190
- device: mi300_1
190
+ device: mi355_dpx
191
timeout_in_minutes: 60
192
parallelism: 2
193
depends_on:
@@ -392,9 +392,9 @@ steps:
392
- if [ "$$BUILDKITE_PARALLEL_JOB" = "2" ]; then pytest -v -s "v1/e2e/spec_decode/acceptance_rates/mtp_other/test_mtp.py::test_gemma4_mtp_acceptance_lengths[True]"; fi
393
mirror:
394
amd:
395
- label: ":amd: (MI355) Spec Decode AL MTP + Other Acceptance Nightly"
395
+ label: ":amd: (MI355 DPX) Spec Decode AL MTP + Other Acceptance Nightly"
396
dind: false
397
- device: mi355_1
397
+ device: mi355_dpx
398
timeout_in_minutes: 55
399
depends_on:
400
- image-build-amd