[ROCm][CI] Purging away redundant test group definitions (#46418)
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Andreas Karatzas committed
Jun 22, 2026 at 17:42 UTC
70ef4d30096bd41e025cdf79c7c89de6ce4e9c79
1 file changed
+131
-283
.buildkite/test-amd.yaml
+131
-283
@@ -161,24 +161,6 @@ steps:
161
commands:
162
- "find compile/ -maxdepth 1 -name 'test_*.py' -print0 | xargs -0 -n1 -I{} pytest -s -v '{}'"
163
164
-- label: PyTorch Fullgraph # TBD
165
- timeout_in_minutes: 180
166
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250]
167
- agent_pool: mi250_1
168
- optional: true
169
- torch_nightly: true
170
- working_dir: "/vllm-workspace/tests"
171
- source_file_dependencies:
172
- - vllm/compilation/
173
- - vllm/model_executor/
174
- - vllm/v1/attention/
175
- - vllm/config/compilation.py
176
- - csrc/
177
- - tests/compile
178
- - vllm/platforms/rocm.py
179
- commands:
180
- - pytest -v -s compile/fullgraph/test_full_graph.py -k 'not test_fp8_kv_scale_compile'
181
-
164
- label: PyTorch Fullgraph Smoke Test # TBD
165
timeout_in_minutes: 180
166
mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250]
@@ -263,37 +245,6 @@ steps:
245
- CUDA_VISIBLE_DEVICES=0,1 pytest -v -s v1/shutdown
246
- pytest -v -s v1/worker/test_worker_memory_snapshot.py
247
266
-- label: Elastic EP Scaling Test # TBD
267
- timeout_in_minutes: 180
268
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250]
269
- agent_pool: mi250_4
270
- num_gpus: 4
271
- working_dir: "/vllm-workspace/tests"
272
- source_file_dependencies:
273
- - vllm/distributed/
274
- - vllm/engine/
275
- - vllm/executor/
276
- - vllm/compilation/
277
- - tests/distributed/
278
- - vllm/platforms/rocm.py
279
- commands:
280
- - pytest -v -s distributed/test_elastic_ep.py
281
-
282
-- label: EPLB Execution # TBD
283
- timeout_in_minutes: 180
284
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250]
285
- agent_pool: mi250_4
286
- num_gpus: 4
287
- working_dir: "/vllm-workspace/tests"
288
- source_file_dependencies:
289
- - vllm/distributed/eplb
290
- - tests/distributed/test_eplb_execute.py
291
- - tests/distributed/test_eplb_spec_decode.py
292
- - vllm/platforms/rocm.py
293
- commands:
294
- - pytest -v -s distributed/test_eplb_execute.py
295
- - pytest -v -s distributed/test_eplb_spec_decode.py
296
-
248
- label: Pipeline + Context Parallelism (4 GPUs) # TBD
249
timeout_in_minutes: 180
250
mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250]
@@ -315,78 +266,8 @@ steps:
266
- pytest -v -s distributed/test_pp_cudagraph.py
267
- pytest -v -s distributed/test_pipeline_parallel.py
268
318
-#----------------------------------------------------------- mi250 · evals -----------------------------------------------------------#
319
-
320
-- label: Multi-Modal Accuracy Eval (Small Models) # TBD
321
- timeout_in_minutes: 180
322
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250]
323
- agent_pool: mi250_1
324
- optional: true
325
- working_dir: "/vllm-workspace/.buildkite/lm-eval-harness"
326
- source_file_dependencies:
327
- - vllm/multimodal/
328
- - vllm/inputs/
329
- - vllm/v1/core/
330
- - vllm/platforms/rocm.py
331
- - vllm/model_executor/model_loader/
332
- commands:
333
- - pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-mm-small.txt --tp-size=1
334
-
335
-#--------------------------------------------------------- mi250 · examples ----------------------------------------------------------#
336
-
337
-- label: Examples # TBD
338
- timeout_in_minutes: 180
339
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250]
340
- agent_pool: mi250_1
341
- optional: true
342
- working_dir: "/vllm-workspace/examples"
343
- source_file_dependencies:
344
- - vllm/entrypoints
345
- - vllm/multimodal
346
- - examples/
347
- - vllm/platforms/rocm.py
348
- commands:
349
- - pip install tensorizer
350
- # Basic
351
- - python3 basic/offline_inference/chat.py --attention-backend TRITON_ATTN
352
- - python3 basic/offline_inference/generate.py --model facebook/opt-125m
353
- - python3 basic/offline_inference/generate.py --model meta-llama/Llama-2-13b-chat-hf --cpu-offload-gb 10
354
- - python3 basic/offline_inference/classify.py
355
- - python3 basic/offline_inference/embed.py
356
- - python3 basic/offline_inference/score.py
357
- # Multi-modal models
358
- - python3 generate/multimodal/audio_language_offline.py --seed 0
359
- - python3 generate/multimodal/vision_language_offline.py --seed 0
360
- - python3 generate/multimodal/vision_language_multi_image_offline.py --seed 0
361
- - python3 generate/multimodal/encoder_decoder_multimodal_offline.py --model-type whisper --seed 0
362
- # Pooling models
363
- - python3 pooling/embed/vision_embedding_offline.py --seed 0
364
- # Features demo
365
- - python3 features/automatic_prefix_caching/prefix_caching_offline.py
366
- - python3 deployment/llm_engine_example.py
367
- - 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
368
- - 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
369
- - 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
370
-
269
#---------------------------------------------------------- mi250 · kernels ----------------------------------------------------------#
270
373
-- label: Kernels Core Operation Test # TBD
374
- timeout_in_minutes: 180
375
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250]
376
- agent_pool: mi250_1
377
- optional: true
378
- working_dir: "/vllm-workspace/tests"
379
- source_file_dependencies:
380
- - csrc/
381
- - tests/kernels/core
382
- - tests/kernels/test_top_k_per_row.py
383
- - tests/kernels/test_concat_mla_q.py
384
- - vllm/model_executor/layers/rotary_embedding/
385
- - vllm/_aiter_ops.py
386
- - vllm/platforms/rocm.py
387
- commands:
388
- - pytest -v -s kernels/core --ignore=kernels/core/test_minimax_reduce_rms.py kernels/test_concat_mla_q.py kernels/test_top_k_per_row.py
389
-
271
- label: Kernels Helion Test # TBD
272
timeout_in_minutes: 180
273
mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250]
@@ -432,50 +313,6 @@ steps:
313
commands:
314
- pytest -v -s models/test_utils.py models/test_vision.py
315
435
-- label: Basic Models Tests (Extra Initialization) %N # TBD
436
- timeout_in_minutes: 180
437
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250]
438
- agent_pool: mi250_1
439
- torch_nightly: true
440
- parallelism: 2
441
- working_dir: "/vllm-workspace/tests"
442
- source_file_dependencies:
443
- - vllm/model_executor/models/
444
- - vllm/model_executor/layers/
445
- - tests/models/test_initialization.py
446
- - tests/models/registry.py
447
- - vllm/_aiter_ops.py
448
- - vllm/platforms/rocm.py
449
- commands:
450
- - pytest -v -s models/test_initialization.py -k 'not test_can_initialize_small_subset' --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
451
-
452
-- label: Basic Models Tests (Initialization) # TBD
453
- timeout_in_minutes: 180
454
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250]
455
- agent_pool: mi250_1
456
- torch_nightly: true
457
- working_dir: "/vllm-workspace/tests"
458
- source_file_dependencies:
459
- - vllm/
460
- - tests/models/test_initialization.py
461
- - tests/models/registry.py
462
- commands:
463
- - pytest -v -s models/test_initialization.py::test_can_initialize_small_subset
464
-
465
-- label: Basic Models Tests (Other) # TBD
466
- timeout_in_minutes: 180
467
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250]
468
- agent_pool: mi250_1
469
- torch_nightly: true
470
- working_dir: "/vllm-workspace/tests"
471
- source_file_dependencies:
472
- - vllm/
473
- - tests/models/test_terratorch.py
474
- - tests/models/test_transformers.py
475
- - tests/models/test_registry.py
476
- commands:
477
- - pytest -v -s models/test_terratorch.py models/test_transformers.py models/test_registry.py
478
-
316
#----------------------------------------------------- mi250 · models / language -----------------------------------------------------#
317
318
- label: Language Models Test (MTEB) # TBD
@@ -500,53 +337,8 @@ steps:
337
commands:
338
- pytest -v -s models/language/generation_ppl_test
339
503
-- label: Language Models Tests (Extra Standard) %N # TBD
504
- timeout_in_minutes: 180
505
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250]
506
- agent_pool: mi250_1
507
- torch_nightly: true
508
- parallelism: 2
509
- working_dir: "/vllm-workspace/tests"
510
- source_file_dependencies:
511
- - vllm/model_executor/models/
512
- - vllm/model_executor/model_loader/
513
- - vllm/model_executor/layers/
514
- - vllm/v1/attention/backends/
515
- - vllm/v1/attention/selector.py
516
- - tests/models/language/pooling/test_embedding.py
517
- - tests/models/language/generation/test_common.py
518
- - tests/models/language/pooling/test_classification.py
519
- - vllm/_aiter_ops.py
520
- - vllm/platforms/rocm.py
521
- commands:
522
- - pip freeze | grep -E 'torch'
523
- - pytest -v -s models/language -m 'core_model and slow_test' --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
524
-
340
#---------------------------------------------------- mi250 · models / multimodal ----------------------------------------------------#
341
527
-- label: Multi-Modal Models (Extended Generation 2) # TBD
528
- timeout_in_minutes: 180
529
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250]
530
- agent_pool: mi250_1
531
- working_dir: "/vllm-workspace/tests"
532
- source_file_dependencies:
533
- - vllm/
534
- - tests/models/multimodal/generation
535
- commands:
536
- - pip install git+https://github.com/TIGER-AI-Lab/Mantis.git
537
- - pytest -v -s models/multimodal/generation/test_common.py -m 'split(group=0) and not core_model'
538
-
539
-- label: Multi-Modal Models (Extended Pooling) # TBD
540
- timeout_in_minutes: 180
541
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250]
542
- agent_pool: mi250_1
543
- working_dir: "/vllm-workspace/tests"
544
- source_file_dependencies:
545
- - vllm/
546
- - tests/models/multimodal/pooling
547
- commands:
548
- - pytest -v -s models/multimodal/pooling -m 'not core_model'
549
-
342
- label: "Multi-Modal Models (Standard) 3: llava + qwen2_vl" # TBD
343
timeout_in_minutes: 180
344
mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250]
@@ -780,21 +572,6 @@ steps:
572
- TP_SIZE=1 DP_SIZE=2 pytest -v -s v1/distributed/test_external_lb_dp.py
573
- DP_SIZE=2 pytest -v -s entrypoints/openai/test_multi_api_servers.py
574
783
-- label: NixlConnector PD + Spec Decode acceptance (2 GPUs) # TBD
784
- timeout_in_minutes: 180
785
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250]
786
- agent_pool: mi250_2
787
- num_gpus: 2
788
- working_dir: "/vllm-workspace/tests"
789
- source_file_dependencies:
790
- - vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py
791
- - vllm/v1/worker/kv_connector_model_runner_mixin.py
792
- - tests/v1/kv_connector/nixl_integration/
793
- - vllm/platforms/rocm.py
794
- commands:
795
- - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt
796
- - ATTENTION_BACKEND=TRITON_ATTN bash v1/kv_connector/nixl_integration/spec_decode_acceptance_test.sh
797
-
575
- label: V1 e2e (2 GPUs) # TBD
576
timeout_in_minutes: 180
577
mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250]
@@ -807,20 +584,6 @@ steps:
584
commands:
585
- pytest -v -s v1/e2e/spec_decode/test_spec_decode.py -k "tensor_parallelism"
586
810
-- label: Distributed NixlConnector PD accuracy (4 GPUs) # TBD
811
- timeout_in_minutes: 180
812
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250]
813
- agent_pool: mi250_4
814
- num_gpus: 4
815
- working_dir: "/vllm-workspace/tests"
816
- source_file_dependencies:
817
- - vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py
818
- - tests/v1/kv_connector/nixl_integration/
819
- - vllm/platforms/rocm.py
820
- commands:
821
- - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt
822
- - ATTENTION_BACKEND=TRITON_ATTN bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh
823
-
587
#------------------------------------------------------------- mi250 · misc ------------------------------------------------------------#
588
589
- label: Async Engine, Inputs, Utils, Worker, Config (CPU) # TBD
@@ -980,6 +743,24 @@ steps:
743
commands:
744
- pytest -s -v compile/passes --ignore compile/passes/distributed
745
746
+- label: PyTorch Fullgraph # TBD
747
+ timeout_in_minutes: 180
748
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
749
+ agent_pool: mi300_1
750
+ optional: true
751
+ torch_nightly: true
752
+ working_dir: "/vllm-workspace/tests"
753
+ source_file_dependencies:
754
+ - vllm/compilation/
755
+ - vllm/model_executor/
756
+ - vllm/v1/attention/
757
+ - vllm/config/compilation.py
758
+ - csrc/
759
+ - tests/compile
760
+ - vllm/platforms/rocm.py
761
+ commands:
762
+ - pytest -v -s compile/fullgraph/test_full_graph.py -k 'not test_fp8_kv_scale_compile'
763
+
764
- label: Pytorch Nightly Dependency Override Check # TBD
765
timeout_in_minutes: 180
766
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
@@ -1057,6 +838,21 @@ steps:
838
- pytest -v -s distributed/test_eplb_algo.py
839
- pytest -v -s distributed/test_eplb_utils.py
840
841
+- label: EPLB Execution # TBD
842
+ timeout_in_minutes: 180
843
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
844
+ agent_pool: mi300_4
845
+ num_gpus: 4
846
+ working_dir: "/vllm-workspace/tests"
847
+ source_file_dependencies:
848
+ - vllm/distributed/eplb
849
+ - tests/distributed/test_eplb_execute.py
850
+ - tests/distributed/test_eplb_spec_decode.py
851
+ - vllm/platforms/rocm.py
852
+ commands:
853
+ - pytest -v -s distributed/test_eplb_execute.py
854
+ - pytest -v -s distributed/test_eplb_spec_decode.py
855
+
856
- label: Distributed Tests (2xH100-2xMI250) # TBD
857
timeout_in_minutes: 180
858
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
@@ -1410,6 +1206,21 @@ steps:
1206
commands:
1207
- pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-small-rocm.txt
1208
1209
+- label: Multi-Modal Accuracy Eval (Small Models) # TBD
1210
+ timeout_in_minutes: 180
1211
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1212
+ agent_pool: mi300_1
1213
+ optional: true
1214
+ working_dir: "/vllm-workspace/.buildkite/lm-eval-harness"
1215
+ source_file_dependencies:
1216
+ - vllm/multimodal/
1217
+ - vllm/inputs/
1218
+ - vllm/v1/core/
1219
+ - vllm/platforms/rocm.py
1220
+ - vllm/model_executor/model_loader/
1221
+ commands:
1222
+ - pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-mm-small.txt --tp-size=1
1223
+
1224
- label: GPQA Eval (GPT-OSS) (2xH100-2xMI300) # TBD
1225
timeout_in_minutes: 180
1226
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
@@ -1742,6 +1553,52 @@ steps:
1553
- pytest -v -s model_executor -m '(not slow_test)'
1554
- pytest -v -s entrypoints/openai/completion/test_tensorizer_entrypoint.py
1555
1556
+#------------------------------------------------------ mi300 · models / basic -------------------------------------------------------#
1557
+
1558
+- label: Basic Models Tests (Extra Initialization) %N # TBD
1559
+ timeout_in_minutes: 180
1560
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1561
+ agent_pool: mi300_1
1562
+ torch_nightly: true
1563
+ parallelism: 2
1564
+ working_dir: "/vllm-workspace/tests"
1565
+ source_file_dependencies:
1566
+ - vllm/model_executor/models/
1567
+ - vllm/model_executor/layers/
1568
+ - tests/models/test_initialization.py
1569
+ - tests/models/registry.py
1570
+ - vllm/_aiter_ops.py
1571
+ - vllm/platforms/rocm.py
1572
+ commands:
1573
+ - pytest -v -s models/test_initialization.py -k 'not test_can_initialize_small_subset' --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
1574
+
1575
+- label: Basic Models Tests (Initialization) # TBD
1576
+ timeout_in_minutes: 180
1577
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1578
+ agent_pool: mi300_1
1579
+ torch_nightly: true
1580
+ working_dir: "/vllm-workspace/tests"
1581
+ source_file_dependencies:
1582
+ - vllm/
1583
+ - tests/models/test_initialization.py
1584
+ - tests/models/registry.py
1585
+ commands:
1586
+ - pytest -v -s models/test_initialization.py::test_can_initialize_small_subset
1587
+
1588
+- label: Basic Models Tests (Other) # TBD
1589
+ timeout_in_minutes: 180
1590
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1591
+ agent_pool: mi300_1
1592
+ torch_nightly: true
1593
+ working_dir: "/vllm-workspace/tests"
1594
+ source_file_dependencies:
1595
+ - vllm/
1596
+ - tests/models/test_terratorch.py
1597
+ - tests/models/test_transformers.py
1598
+ - tests/models/test_registry.py
1599
+ commands:
1600
+ - pytest -v -s models/test_terratorch.py models/test_transformers.py models/test_registry.py
1601
+
1602
#----------------------------------------------------- mi300 · models / language -----------------------------------------------------#
1603
1604
- label: Language Models Test (Extended Pooling) # TBD
@@ -1770,6 +1627,28 @@ steps:
1627
- pip freeze | grep -E 'torch'
1628
- pytest -v -s models/language -m 'core_model and (not slow_test)'
1629
1630
+- label: Language Models Tests (Extra Standard) %N # TBD
1631
+ timeout_in_minutes: 180
1632
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
1633
+ agent_pool: mi300_1
1634
+ torch_nightly: true
1635
+ parallelism: 2
1636
+ working_dir: "/vllm-workspace/tests"
1637
+ source_file_dependencies:
1638
+ - vllm/model_executor/models/
1639
+ - vllm/model_executor/model_loader/
1640
+ - vllm/model_executor/layers/
1641
+ - vllm/v1/attention/backends/
1642
+ - vllm/v1/attention/selector.py
1643
+ - tests/models/language/pooling/test_embedding.py
1644
+ - tests/models/language/generation/test_common.py
1645
+ - tests/models/language/pooling/test_classification.py
1646
+ - vllm/_aiter_ops.py
1647
+ - vllm/platforms/rocm.py
1648
+ commands:
1649
+ - pip freeze | grep -E 'torch'
1650
+ - pytest -v -s models/language -m 'core_model and slow_test' --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
1651
+
1652
#---------------------------------------------------- mi300 · models / multimodal ----------------------------------------------------#
1653
1654
- label: Multi-Modal Models (Extended Generation 1) # TBD
@@ -2336,6 +2215,21 @@ steps:
2215
- TP_SIZE=1 DP_SIZE=2 pytest -v -s v1/distributed/test_external_lb_dp.py
2216
- DP_SIZE=2 pytest -v -s entrypoints/openai/test_multi_api_servers.py
2217
2218
+- label: NixlConnector PD + Spec Decode acceptance (2 GPUs) # TBD
2219
+ timeout_in_minutes: 180
2220
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
2221
+ agent_pool: mi300_2
2222
+ num_gpus: 2
2223
+ working_dir: "/vllm-workspace/tests"
2224
+ source_file_dependencies:
2225
+ - vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py
2226
+ - vllm/v1/worker/kv_connector_model_runner_mixin.py
2227
+ - tests/v1/kv_connector/nixl_integration/
2228
+ - vllm/platforms/rocm.py
2229
+ commands:
2230
+ - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt
2231
+ - ATTENTION_BACKEND=TRITON_ATTN bash v1/kv_connector/nixl_integration/spec_decode_acceptance_test.sh
2232
+
2233
- label: Distributed Tests (2xH100-2xMI300) # TBD
2234
timeout_in_minutes: 180
2235
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
@@ -3411,52 +3305,6 @@ steps:
3305
commands:
3306
- pytest -v -s -m 'not slow_test' v1/spec_decode
3307
3414
-- label: NixlConnector PD + Spec Decode acceptance (2 GPUs) # TBD
3415
- timeout_in_minutes: 180
3416
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
3417
- agent_pool: mi355_2
3418
- num_gpus: 2
3419
- optional: true
3420
- working_dir: "/vllm-workspace/tests"
3421
- source_file_dependencies:
3422
- - vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py
3423
- - vllm/v1/worker/kv_connector_model_runner_mixin.py
3424
- - tests/v1/kv_connector/nixl_integration/
3425
- - vllm/platforms/rocm.py
3426
- commands:
3427
- - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt
3428
- - ATTENTION_BACKEND=TRITON_ATTN bash v1/kv_connector/nixl_integration/spec_decode_acceptance_test.sh
3429
-
3430
-- label: Distributed NixlConnector PD accuracy (4 GPUs) # TBD
3431
- timeout_in_minutes: 180
3432
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
3433
- agent_pool: mi355_4
3434
- num_gpus: 4
3435
- optional: true
3436
- working_dir: "/vllm-workspace/tests"
3437
- source_file_dependencies:
3438
- - vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py
3439
- - tests/v1/kv_connector/nixl_integration/
3440
- - vllm/platforms/rocm.py
3441
- commands:
3442
- - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt
3443
- - ATTENTION_BACKEND=TRITON_ATTN bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh
3444
-
3445
-- label: DP EP Distributed NixlConnector PD accuracy tests (4 GPUs) # TBD
3446
- timeout_in_minutes: 180
3447
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
3448
- agent_pool: mi355_4
3449
- num_gpus: 4
3450
- optional: true
3451
- working_dir: "/vllm-workspace/tests"
3452
- source_file_dependencies:
3453
- - vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py
3454
- - tests/v1/kv_connector/nixl_integration/
3455
- - vllm/platforms/rocm.py
3456
- commands:
3457
- - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt
3458
- - DP_EP=1 ATTENTION_BACKEND=TRITON_ATTN bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh
3459
-
3308
#------------------------------------------------------ mi355 · weight_loading -------------------------------------------------------#
3309
3310
- label: Weight Loading Multiple GPU # TBD