[ROCm][CI] Extend AMD coverage for distributed, model, and eval tests (#56679)

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> Signed-off-by: Stefan Koncarevic <stefan.koncarevic@amd.com> Co-authored-by: OpenAI Codex <codex@openai.com> Co-authored-by: Stefan Koncarevic <stefan.koncarevic@amd.com>
2026-10-02 23:20:51 -05:00
parent 28556f8600
commit c28834df00
6 changed files with 449 additions and 0 deletions
@@ -93,6 +93,24 @@ steps:
  commands:
  - export VLLM_TEST_CLEAN_GPU_MEMORY=1
  - pytest -s -v tests/compile/passes/distributed
+  mirror:
+    amd:
+      label: ':amd: (MI300) Distributed Compile'
+      dind: false
+      device: mi300_2
+      timeout_in_minutes: 65
+      working_dir: /vllm-workspace/
+      depends_on:
+      - image-build-amd
+      source_file_dependencies:
+      - vllm/compilation/
+      - vllm/model_executor/layers
+      - tests/compile/passes/distributed/
+      - vllm/_aiter_ops.py
+      - vllm/platforms/rocm.py
+      commands:
+      - export VLLM_TEST_CLEAN_GPU_MEMORY=1
+      - pytest -s -v tests/compile/passes/distributed

 - label: ":nvidia: (B200) Fusion and Compile"
  key: fusion-and-compile-unit-tests-2xb200
@@ -463,3 +481,45 @@ steps:
      - amd-smi
      - VLLM_ROCM_USE_AITER=1 VLLM_ROCM_USE_AITER_CUSTOM_AR=1 pytest -v -s tests/compile/fusions_e2e/test_tp2_ar_rms.py -k 'test_tp2_ar_rms_fp8_fusions and qwen3 and ROCM_AITER_UNIFIED_ATTN and +quant_fp8 and (dynamo_partition or not +rms_norm)'
      - pytest -v -s tests/compile/fusions_e2e/test_tp2_rocm_moe.py
+
+- label: ':amd: (MI355) Fusion E2E TP2 AR-RMS (Inductor Partition)'
+  key: fusion-e2e-tp2-ar-rms-inductor-partition-amd
+  dind: false
+  device: mi355_2
+  num_devices: 2
+  timeout_in_minutes: 65
+  working_dir: /vllm-workspace/
+  depends_on:
+  - image-build-amd
+  source_file_dependencies:
+  - vllm/compilation/
+  - vllm/model_executor/layers
+  - tests/compile/fusions_e2e/
+  - vllm/_aiter_ops.py
+  - vllm/platforms/rocm.py
+  - csrc/quantization/
+  - vllm/v1/attention/
+  commands:
+  - export VLLM_TEST_CLEAN_GPU_MEMORY=1
+  - pytest -v -s tests/compile/fusions_e2e/test_tp2_ar_rms.py::test_tp2_ar_rms_fusions -k "inductor_partition"
+
+- label: ':amd: (MI355) Fusion E2E TP2 AR-RMS (Dynamo Partition)'
+  key: fusion-e2e-tp2-ar-rms-dynamo-partition-amd
+  dind: false
+  device: mi355_2
+  num_devices: 2
+  timeout_in_minutes: 65
+  working_dir: /vllm-workspace/
+  depends_on:
+  - image-build-amd
+  source_file_dependencies:
+  - vllm/compilation/
+  - vllm/model_executor/layers
+  - tests/compile/fusions_e2e/
+  - vllm/_aiter_ops.py
+  - vllm/platforms/rocm.py
+  - csrc/quantization/
+  - vllm/v1/attention/
+  commands:
+  - export VLLM_TEST_CLEAN_GPU_MEMORY=1
+  - pytest -v -s tests/compile/fusions_e2e/test_tp2_ar_rms.py::test_tp2_ar_rms_fusions -k "dynamo_partition"
@@ -17,6 +17,24 @@ steps:
  - pytest -v -s distributed/test_shm_broadcast.py
  - pytest -v -s distributed/test_shm_buffer.py
  - pytest -v -s distributed/test_shm_storage.py
+  mirror:
+    amd:
+      label: ':amd: (MI300) Distributed Comm Ops'
+      dind: false
+      device: mi300_2
+      timeout_in_minutes: 30
+      working_dir: /vllm-workspace/tests
+      depends_on:
+      - image-build-amd
+      source_file_dependencies:
+      - vllm/distributed
+      - tests/distributed
+      - vllm/platforms/rocm.py
+      commands:
+      - pytest -v -s distributed/test_comm_ops.py
+      - pytest -v -s distributed/test_shm_broadcast.py
+      - pytest -v -s distributed/test_shm_buffer.py
+      - pytest -v -s distributed/test_shm_storage.py

 - label: ":nvidia: (L4) Sharded RDT Weight Transfer"
  key: sharded-rdt-weight-transfer
@@ -176,6 +194,31 @@ steps:
  - VLLM_TEST_SAME_HOST=1 VLLM_TEST_WITH_DEFAULT_DEVICE_SET=1 torchrun --nproc-per-node=4 distributed/test_same_node.py | grep 'Same node test passed'
  - CUDA_VISIBLE_DEVICES=0,1 pytest -v -s v1/shutdown
  - pytest -v -s v1/worker/test_worker_memory_snapshot.py
+  mirror:
+    amd:
+      label: ':amd: (MI300) Distributed Torchrun + Shutdown'
+      dind: false
+      device: mi300_2
+      timeout_in_minutes: 35
+      working_dir: /vllm-workspace/tests
+      depends_on:
+      - image-build-amd
+      source_file_dependencies:
+      - vllm/distributed/
+      - vllm/engine/
+      - vllm/v1/executor/
+      - vllm/v1/worker/worker_base.py
+      - vllm/v1/engine/
+      - vllm/v1/worker/
+      - tests/distributed/
+      - tests/v1/shutdown
+      - tests/v1/worker/test_worker_memory_snapshot.py
+      - vllm/platforms/rocm.py
+      commands:
+      - VLLM_TEST_SAME_HOST=1 torchrun --nproc-per-node=4 distributed/test_same_node.py | grep 'Same node test passed'
+      - VLLM_TEST_SAME_HOST=1 VLLM_TEST_WITH_DEFAULT_DEVICE_SET=1 torchrun --nproc-per-node=4 distributed/test_same_node.py | grep 'Same node test passed'
+      - HIP_VISIBLE_DEVICES=0,1 pytest -v -s v1/shutdown
+      - pytest -v -s v1/worker/test_worker_memory_snapshot.py

 - label: ":nvidia: (L4) Distributed Torchrun + Examples"
  key: distributed-torchrun-examples-4-gpus
@@ -282,6 +325,31 @@ steps:
  - pytest -v -s v1/engine/test_engine_core_client.py::test_kv_cache_events_dp
  - pytest -v -s distributed/test_utils.py
  - pytest -v -s distributed/test_engram_dp_shard.py
+  mirror:
+    amd:
+      label: ':amd: (MI355) Distributed DP Extended'
+      dind: false
+      device: mi355_4
+      timeout_in_minutes: 60
+      working_dir: /vllm-workspace/tests
+      depends_on:
+      - image-build-amd
+      source_file_dependencies:
+      - vllm/distributed/
+      - tests/v1/distributed
+      - tests/v1/engine/test_engine_core_client.py
+      - tests/distributed/test_utils
+      - vllm/platforms/rocm.py
+      commands:
+      - export NCCL_CUMEM_HOST_ENABLE=0
+      - TP_SIZE=2 DP_SIZE=2 pytest -v -s v1/distributed/test_async_llm_dp.py
+      - TP_SIZE=2 DP_SIZE=2 pytest -v -s v1/distributed/test_eagle_dp.py
+      - TP_SIZE=2 DP_SIZE=2 pytest -v -s v1/distributed/test_external_lb_dp.py
+      - TP_SIZE=1 DP_SIZE=4 pytest -v -s v1/distributed/test_dense_dp_world_size.py
+      - TP_SIZE=1 DP_SIZE=4 pytest -v -s v1/distributed/test_internal_lb_dp.py
+      - TP_SIZE=1 DP_SIZE=4 pytest -v -s v1/distributed/test_hybrid_lb_dp.py
+      - pytest -v -s v1/engine/test_engine_core_client.py::test_kv_cache_events_dp
+      - pytest -v -s distributed/test_utils.py

 - label: ":nvidia: (L4) Distributed Compile + Comm"
  key: distributed-compile-comm-4-gpus
@@ -304,6 +372,27 @@ steps:
  - pytest -v -s distributed/test_symm_mem_allreduce.py
  # test multi-node TP with multiproc executor (simulated on single node)
  - pytest -v -s distributed/test_multiproc_executor.py::test_multiproc_executor_multi_node
+  mirror:
+    amd:
+      label: ':amd: (MI355) Distributed Compile + Comm'
+      dind: false
+      device: mi355_4
+      timeout_in_minutes: 45
+      working_dir: /vllm-workspace/tests
+      depends_on:
+      - image-build-amd
+      source_file_dependencies:
+      - vllm/distributed/
+      - tests/distributed/test_pynccl
+      - tests/distributed/test_events
+      - tests/distributed/test_symm_mem_allreduce.py
+      - tests/distributed/test_multiproc_executor.py
+      - vllm/platforms/rocm.py
+      commands:
+      - pytest -v -s distributed/test_pynccl.py
+      - pytest -v -s distributed/test_events.py
+      - pytest -v -s distributed/test_symm_mem_allreduce.py
+      - pytest -v -s distributed/test_multiproc_executor.py::test_multiproc_executor_multi_node

 - label: ":nvidia: (H100) Distributed DP + EP"
  key: distributed-tests-8xh100
@@ -359,6 +448,28 @@ steps:
  - torchrun --nproc_per_node=2 distributed/test_ca_buffer_sharing.py
  - TARGET_TEST_SUITE=A100 pytest basic_correctness/ -v -s -m 'distributed(num_gpus=2)'
  - pytest -v -s -x lora/test_mixtral.py
+  mirror:
+    amd:
+      label: ':amd: (MI355) Distributed'
+      dind: false
+      device: mi355_4
+      timeout_in_minutes: 40
+      working_dir: /vllm-workspace/tests
+      depends_on:
+      - image-build-amd
+      source_file_dependencies:
+      - vllm/
+      - tests/models/test_vision.py
+      - vllm/platforms/rocm.py
+      - tests/distributed/
+      - tests/basic_correctness/
+      - tests/lora/test_mixtral.py
+      commands:
+      - pytest -v -s distributed/test_custom_all_reduce.py
+      - torchrun --nproc_per_node=2 distributed/test_ca_buffer_sharing.py
+      - TARGET_TEST_SUITE=MI355 pytest basic_correctness/ -v -s -m 'distributed(num_gpus=2)'
+      - pytest -v -s -x lora/test_mixtral.py
+      - pytest -v -s models/test_vision.py -m 'distributed(num_gpus=4)'

 - label: ":nvidia: (H100) Distributed Features"
  key: distributed-tests-2xh100-2xmi300
@@ -583,3 +694,27 @@ steps:
  - pytest -v -s distributed/test_ray_v2_executor_e2e.py
  - pytest -v -s distributed/test_pipeline_parallel.py -k "ray"
  - TARGET_TEST_SUITE=L4 pytest -v -s basic_correctness/models/test_basic_correctness.py -k "ray"
+  mirror:
+    amd:
+      label: ':amd: (MI355) RayExecutorV2'
+      dind: false
+      device: mi355_4
+      timeout_in_minutes: 75
+      working_dir: /vllm-workspace/tests
+      depends_on:
+      - image-build-amd
+      source_file_dependencies:
+      - vllm/v1/executor/ray_executor_v2.py
+      - vllm/v1/executor/abstract.py
+      - vllm/v1/executor/multiproc_executor.py
+      - tests/distributed/test_ray_v2_executor.py
+      - tests/distributed/test_ray_v2_executor_e2e.py
+      - tests/distributed/test_pipeline_parallel.py
+      - tests/basic_correctness/models/test_basic_correctness.py
+      - vllm/platforms/rocm.py
+      commands:
+      - export VLLM_USE_RAY_V2_EXECUTOR_BACKEND=1
+      - pytest -v -s distributed/test_ray_v2_executor.py
+      - pytest -v -s distributed/test_ray_v2_executor_e2e.py
+      - pytest -v -s distributed/test_pipeline_parallel.py -k "ray"
+      - TARGET_TEST_SUITE=MI355 pytest -v -s basic_correctness/models/test_basic_correctness.py -k "ray"
@@ -150,3 +150,27 @@ steps:
      - .buildkite/scripts/scheduled_integration_test/
      - vllm/_aiter_ops.py
      - vllm/platforms/rocm.py
+
+- label: ":amd: (MI355) Qwen3-Next-80B-A3B-Instruct MTP Async EPLB Accuracy"
+  key: amd-qwen3-next-mtp-async-eplb-accuracy
+  dind: false
+  device: mi355_4
+  timeout_in_minutes: 45
+  depends_on:
+  - image-build-amd
+  working_dir: /vllm-workspace
+  source_file_dependencies:
+  - vllm/model_executor/models/
+  - vllm/model_executor/model_loader/
+  - vllm/v1/spec_decode/
+  - vllm/distributed/eplb
+  - vllm/model_executor/layers/fused_moe/
+  - vllm/model_executor/layers/quantization/
+  - vllm/v1/attention/backends/
+  - vllm/v1/attention/selector.py
+  - .buildkite/scripts/scheduled_integration_test/
+  - vllm/_aiter_ops.py
+  - vllm/platforms/rocm.py
+  num_devices: 4
+  commands:
+  - VLLM_ENGINE_READY_TIMEOUT_S=1800 bash .buildkite/scripts/scheduled_integration_test/qwen3_next_mtp_async_eplb.sh 0.8 1319 8040
@@ -753,6 +753,32 @@ steps:
  parallelism: 4
  commands:
    - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/moe-refactor/config-b200-shard-$$BUILDKITE_PARALLEL_JOB.txt
+  mirror:
+    amd:
+      label: ":amd: (MI355) MoE Refactor BF16 Integration"
+      dind: false
+      device: mi355_2
+      num_devices: 2
+      parallelism: 1
+      timeout_in_minutes: 105
+      working_dir: /vllm-workspace/tests
+      depends_on:
+      - image-build-amd
+      source_file_dependencies:
+      - tests/evals/gsm8k/
+      - tests/utils.py
+      - vllm/model_executor/layers/fused_moe/
+      - vllm/model_executor/models/
+      - vllm/model_executor/model_loader/
+      - vllm/model_executor/kernels/linear/
+      - vllm/v1/attention/
+      - vllm/distributed/
+      - vllm/_aiter_ops.py
+      - vllm/platforms/rocm.py
+      env:
+        VLLM_WORKER_MULTIPROC_METHOD: "spawn"
+      commands:
+      - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/moe-refactor/config-b200.txt -k "Llama-4-Scout-BF16-triton or Mixtral-8x7B-BF16-triton"

 - label: ":nvidia: (B200) MoE Refactor DP Integration TEMPORARY"
  key: moe-refactor-integration-test-b200-dp-temporary
@@ -761,6 +787,32 @@ steps:
  num_devices: 2
  commands:
    - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/moe-refactor-dp-ep/config-b200.txt
+  mirror:
+    amd:
+      label: ":amd: (MI355) MoE Refactor BF16 DP Integration"
+      dind: false
+      device: mi355_2
+      num_devices: 2
+      timeout_in_minutes: 105
+      working_dir: /vllm-workspace/tests
+      depends_on:
+      - image-build-amd
+      source_file_dependencies:
+      - tests/evals/gsm8k/
+      - tests/utils.py
+      - vllm/config/parallel.py
+      - vllm/model_executor/layers/fused_moe/
+      - vllm/model_executor/models/
+      - vllm/model_executor/model_loader/
+      - vllm/model_executor/kernels/linear/
+      - vllm/v1/attention/
+      - vllm/distributed/
+      - vllm/_aiter_ops.py
+      - vllm/platforms/rocm.py
+      env:
+        VLLM_WORKER_MULTIPROC_METHOD: "spawn"
+      commands:
+      - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/moe-refactor-dp-ep/config-b200.txt -k "Qwen3-30B-A3B-BF16-triton"

 - label: ":nvidia: (L4) LM Eval Humming TEMPORARY"
  key: lm-eval-humming-l4
@@ -1068,6 +1120,29 @@ steps:
  - tests/evals/gpt_oss/
  commands:
    - pytest -s -v evals/gpt_oss/test_gpqa_correctness.py --config-list-file=configs/models-h100.txt
+  mirror:
+    amd:
+      label: ":amd: (MI300) GPQA Eval (GPT-OSS)"
+      dind: false
+      device: mi300_2
+      num_devices: 2
+      timeout_in_minutes: 40
+      working_dir: /vllm-workspace/tests
+      depends_on:
+      - image-build-amd
+      source_file_dependencies:
+      - csrc/
+      - vllm/model_executor/layers/quantization
+      - vllm/model_executor/models/
+      - vllm/model_executor/model_loader/
+      - vllm/v1/attention/backends/
+      - vllm/v1/attention/selector.py
+      - vllm/model_executor/layers/fused_moe/
+      - tests/evals/gpt_oss/
+      - vllm/_aiter_ops.py
+      - vllm/platforms/rocm.py
+      commands:
+      - pytest -s -v evals/gpt_oss/test_gpqa_correctness.py --config-list-file=configs/models-gfx942.txt

 - label: ":nvidia: (B200) GPQA Eval (GPT-OSS)"
  key: gpqa-eval-gpt-oss-2xb200
@@ -1176,6 +1251,28 @@ steps:
  - tests/evals/gpt_oss/
  commands:
    - pytest -s -v evals/gpt_oss/test_gpqa_correctness.py --config-list-file=configs/models-spark.txt
+  mirror:
+    amd:
+      label: ":amd: (MI355 DPX) GPQA Eval (GPT-OSS TP1)"
+      # The Spark config is a backend-neutral GPT-OSS TP1 baseline.
+      dind: false
+      device: mi355_dpx
+      num_devices: 1
+      timeout_in_minutes: 40
+      working_dir: /vllm-workspace/tests
+      depends_on:
+      - image-build-amd
+      source_file_dependencies:
+      - csrc/
+      - vllm/model_executor/layers/quantization
+      - vllm/model_executor/models/
+      - vllm/model_executor/model_loader/
+      - vllm/v1/attention/backends/
+      - vllm/v1/attention/selector.py
+      - vllm/model_executor/layers/fused_moe/
+      - tests/evals/gpt_oss/
+      - vllm/_aiter_ops.py
+      - vllm/platforms/rocm.py

 - label: ":nvidia: (H200 MIG 35GB) LM Eval KV-Offload"
  key: kv-offload-small
@@ -1292,3 +1389,53 @@ steps:
      - vllm/_aiter_ops.py
      - vllm/platforms/rocm.py
      - tests/evals/mrcr/
+
+- label: ":amd: (MI355) LM Eval Large Models Harness"
+  key: amd-lm-eval-large-models-harness
+  dind: false
+  device: mi355_8
+  optional: true
+  timeout_in_minutes: 85
+  depends_on:
+  - image-build-amd
+  working_dir: /vllm-workspace/.buildkite/lm-eval-harness
+  source_file_dependencies:
+  - csrc/
+  - vllm/model_executor/layers/quantization
+  - vllm/model_executor/models/
+  - vllm/model_executor/model_loader/
+  - vllm/v1/attention/backends/
+  - vllm/v1/attention/selector.py
+  - vllm/_aiter_ops.py
+  - vllm/platforms/rocm.py
+  - .buildkite/lm-eval-harness/
+  num_devices: 8
+  commands:
+  - export VLLM_WORKER_MULTIPROC_METHOD=spawn
+  # Qwen2-57B has 28 attention heads and cannot run at TP8.
+  - pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-large.txt --tp-size=4
+
+- label: ":amd: (MI300) LM Eval Large Models ROCm Harness"
+  key: amd-lm-eval-large-models-rocm-harness
+  dind: false
+  device: mi300_8
+  optional: true
+  timeout_in_minutes: 40
+  depends_on:
+  - image-build-amd
+  working_dir: /vllm-workspace/.buildkite/lm-eval-harness
+  source_file_dependencies:
+  - vllm/model_executor/models/
+  - vllm/model_executor/model_loader/
+  - vllm/model_executor/layers/quantization/
+  - vllm/v1/attention/backends/
+  - vllm/v1/attention/selector.py
+  - vllm/model_executor/layers/layernorm.py
+  - csrc/
+  - vllm/_aiter_ops.py
+  - vllm/platforms/rocm.py
+  - .buildkite/lm-eval-harness/
+  num_devices: 8
+  commands:
+  - export VLLM_WORKER_MULTIPROC_METHOD=spawn
+  - pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-large-rocm.txt --tp-size=8 --timeout=5400
@@ -259,6 +259,25 @@ steps:
  - tests/compile
  commands:
  - pytest -s -v compile/fullgraph/test_full_cudagraph.py
+  mirror:
+    amd:
+      label: ':amd: (MI250) PyTorch Fullgraph CUDAGraph Compatibility'
+      dind: false
+      device: mi250_1
+      timeout_in_minutes: 65
+      working_dir: /vllm-workspace/tests
+      depends_on:
+      - image-build-amd
+      source_file_dependencies:
+      - vllm/compilation/
+      - vllm/model_executor/
+      - vllm/v1/attention/
+      - vllm/config/compilation.py
+      - csrc/
+      - tests/compile
+      - vllm/platforms/rocm.py
+      commands:
+      - pytest -s -v compile/fullgraph/test_full_cudagraph.py

 - label: ":nvidia: (H200 MIG 18GB) PyTorch Nightly Dependency Override Check"
  key: pytorch-nightly-dependency-override-check
@@ -97,6 +97,27 @@ steps:
    - tests/v1/e2e/spec_decode/
  commands:
    - pytest -v -s v1/e2e/spec_decode/eagle/
+  mirror:
+    amd:
+      label: ":amd: (MI355) Spec Decode Eagle Nightly"
+      dind: false
+      device: mi355_1
+      num_devices: 1
+      timeout_in_minutes: 120
+      working_dir: /vllm-workspace/tests
+      depends_on:
+      - image-build-amd
+      source_file_dependencies:
+      - vllm/v1/spec_decode/
+      - vllm/v1/worker/gpu/spec_decode/
+      - vllm/model_executor/models/
+      - vllm/model_executor/model_loader/
+      - vllm/v1/sample/
+      - vllm/model_executor/layers/
+      - vllm/v1/attention/
+      - tests/v1/e2e/spec_decode/
+      - vllm/_aiter_ops.py
+      - vllm/platforms/rocm.py

 # MTP correctness is one parametrized test split into per-model folders;
 # each job below runs whole directories.
@@ -234,6 +255,28 @@ steps:
  commands:
    - pytest -v -s v1/e2e/spec_decode/speculators/
    - pytest -v -s v1/e2e/spec_decode/mtp/
+  mirror:
+    amd:
+      label: ":amd: (MI355) Spec Decode Speculators + MTP Nightly"
+      dind: false
+      device: mi355_1
+      num_devices: 1
+      timeout_in_minutes: 90
+      working_dir: /vllm-workspace/tests
+      depends_on:
+      - image-build-amd
+      source_file_dependencies:
+      - vllm/v1/spec_decode/
+      - vllm/v1/worker/gpu/spec_decode/
+      - vllm/model_executor/models/
+      - vllm/model_executor/model_loader/
+      - vllm/v1/sample/
+      - vllm/model_executor/layers/
+      - vllm/v1/attention/
+      - vllm/transformers_utils/configs/speculators/
+      - tests/v1/e2e/spec_decode/
+      - vllm/_aiter_ops.py
+      - vllm/platforms/rocm.py

 - label: ":nvidia: (H200 MIG 35GB) Spec Decode N-Gram + Suffix"
  key: spec-decode-ngram-suffix
@@ -304,6 +347,27 @@ steps:
    - tests/v1/e2e/spec_decode/
  commands:
    - pytest -v -s v1/e2e/spec_decode/draft_model/
+  mirror:
+    amd:
+      label: ":amd: (MI355) Spec Decode Draft Model Nightly"
+      dind: false
+      device: mi355_1
+      num_devices: 1
+      timeout_in_minutes: 120
+      working_dir: /vllm-workspace/tests
+      depends_on:
+      - image-build-amd
+      source_file_dependencies:
+      - vllm/v1/spec_decode/
+      - vllm/v1/worker/gpu/spec_decode/
+      - vllm/model_executor/models/
+      - vllm/model_executor/model_loader/
+      - vllm/v1/sample/
+      - vllm/model_executor/layers/
+      - vllm/v1/attention/
+      - tests/v1/e2e/spec_decode/
+      - vllm/_aiter_ops.py
+      - vllm/platforms/rocm.py

 - label: ":nvidia: (H100) Speculators Correctness Nightly"
  key: speculators-correctness