Mirrored test group definitions for AMD (2025-11-11) (#28573)

Signed-off-by: Alexei V. Ivanov <alexei.ivanov@amd.com>

.buildkite/test-amd.yaml

@@ -226,6 +226,27 @@ steps:
   - VLLM_ALLOW_INSECURE_SERIALIZATION=1 RAY_DEDUP_LOGS=0 python3 rlhf_colocate.py
   - popd
 
+- label: Distributed Tests (8 GPUs) # 4min
+  timeout_in_minutes: 10
+  mirror_hardwares: [amdexperimental]
+  agent_pool: mi325_8
+  # grade: Blocking
+  gpu: h100
+  num_gpus: 8
+  working_dir: "/vllm-workspace/tests"
+  source_file_dependencies:
+  - examples/offline_inference/torchrun_dp_example.py
+  - vllm/config/parallel.py
+  - vllm/distributed/
+  - vllm/v1/engine/llm_engine.py
+  - vllm/v1/executor/uniproc_executor.py
+  - vllm/v1/worker/gpu_worker.py
+  commands:
+  # https://github.com/NVIDIA/nccl/issues/1838
+  #- export NCCL_CUMEM_HOST_ENABLE=0
+  # test with torchrun tp=2 and dp=4 with ep
+  - torchrun --nproc-per-node=8 ../examples/offline_inference/torchrun_dp_example.py --tp-size=2 --pp-size=1 --dp-size=4 --enable-ep
+
 - label: EPLB Algorithm Test # 5min
   mirror_hardwares: [amdexperimental, amdproduction]
   agent_pool: mi325_1
@@ -238,11 +259,11 @@ steps:
   commands:
   - pytest -v -s distributed/test_eplb_algo.py
 
-- label: EPLB Execution Test # 5min
+- label: EPLB Execution Test # 10min
   mirror_hardwares: [amdexperimental, amdproduction]
   agent_pool: mi325_4
   # grade: Blocking
-  timeout_in_minutes: 15
+  timeout_in_minutes: 20
   working_dir: "/vllm-workspace/tests"
   num_gpus: 4
   source_file_dependencies:
@@ -250,6 +271,7 @@ steps:
   - tests/distributed/test_eplb_execute.py
   commands:
   - pytest -v -s distributed/test_eplb_execute.py
+  - pytest -v -s distributed/test_eplb_spec_decode.py
 
 - label: Metrics, Tracing Test # 12min
   timeout_in_minutes: 20
@@ -273,7 +295,7 @@ steps:
 
 - label: Regression Test # 7min
   timeout_in_minutes: 20
-  mirror_hardwares: [amdexperimental, amdproduction]
+  mirror_hardwares: [amdexperimental, amdproduction, amdtentative]
   agent_pool: mi325_1
   grade: Blocking
   source_file_dependencies:
@@ -288,7 +310,7 @@ steps:
   timeout_in_minutes: 40
   mirror_hardwares: [amdexperimental, amdproduction]
   agent_pool: mi325_1
-  #grade: Blocking
+  # grade: Blocking
   source_file_dependencies:
   - vllm/
   - tests/engine
@@ -337,6 +359,7 @@ steps:
     - tests/v1
   commands:
     # split the test to avoid interference
+    - uv pip install --system -r /vllm-workspace/requirements/kv_connectors.txt
     - pytest -v -s -m 'not cpu_test' v1/core
     - pytest -v -s v1/executor
     - pytest -v -s v1/kv_offload
@@ -344,7 +367,7 @@ steps:
     - pytest -v -s v1/logits_processors
     - pytest -v -s v1/worker
     - pytest -v -s v1/spec_decode
-    - pytest -v -s -m 'not cpu_test' v1/kv_connector/unit --ignore=v1/kv_connector/unit/test_lmcache_integration.py
+    - pytest -v -s -m 'not cpu_test' v1/kv_connector/unit
     - pytest -v -s -m 'not cpu_test' v1/metrics
     - pytest -v -s v1/test_oracle.py
     - pytest -v -s v1/test_request.py
@@ -353,6 +376,20 @@ steps:
     - pip install -U git+https://github.com/robertgshaw2-redhat/lm-evaluation-harness.git@streaming-api
     - pytest -v -s entrypoints/openai/correctness/test_lmeval.py::test_lm_eval_accuracy_v1_engine
 
+# TODO: Add the "V1 Test attetion (MI300)" test group
+
+- label: V1 Test attention (H100) # 10min
+  mirror_hardwares: [amdexperimental]
+  agent_pool: mi325_1
+  # grade: Blocking
+  timeout_in_minutes: 30
+  gpu: h100
+  source_file_dependencies:
+    - vllm/v1/attention
+    - tests/v1/attention
+  commands:
+    - pytest -v -s v1/attention
+
 - label: V1 Test others (CPU) # 5 mins
   mirror_hardwares: [amdexperimental, amdproduction]
   agent_pool: mi325_1
@@ -479,10 +516,11 @@ steps:
   - tests/compile
   commands:
   - pytest -v -s compile/test_basic_correctness.py
+  - pytest -v -s compile/test_multimodal_compile.py
   - pytest -v -s compile/piecewise/
 
-- label: PyTorch Fullgraph Test # 22min
-  timeout_in_minutes: 35
+- label: PyTorch Fullgraph Test # 27min
+  timeout_in_minutes: 40
   mirror_hardwares: [amdexperimental, amdproduction]
   agent_pool: mi325_1
   # grade: Blocking
@@ -491,8 +529,23 @@ steps:
   - vllm/
   - tests/compile
   commands:
-  - pytest -v -s compile/test_full_graph.py
-  - pytest -v -s compile/test_fusions_e2e.py
+  - pytest -v -s compile/test_full_graph.py -k 'not test_fp8_kv_scale_compile'
+    # Limit to no custom ops to reduce running time
+    # Wrap with quotes to escape yaml and avoid starting -k string with a -
+  - "pytest -v -s compile/test_fusions_e2e.py -k 'TRITON and -quant_fp8'"
+
+- label: Cudagraph test
+  timeout_in_minutes: 20
+  mirror_hardwares: [amdexperimental, amdproduction]
+  agent_pool: mi325_1
+  source_file_dependencies:
+  - tests/v1/cudagraph
+  - vllm/v1/cudagraph_dispatcher.py
+  - vllm/config/compilation.py
+  - vllm/compilation
+  commands:
+    - pytest -v -s v1/cudagraph/test_cudagraph_dispatch.py
+    - pytest -v -s v1/cudagraph/test_cudagraph_mode.py
 
 - label: Kernels Core Operation Test # 48min
   timeout_in_minutes: 75
@@ -544,6 +597,8 @@ steps:
   - tests/kernels/moe
   - vllm/model_executor/layers/fused_moe/
   - vllm/distributed/device_communicators/
+  - vllm/envs.py
+  - vllm/config
   commands:
     - pytest -v -s kernels/moe --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
   parallelism: 2
@@ -562,10 +617,13 @@ steps:
 
 - label: Model Executor Test # 23min
   timeout_in_minutes: 35
+  torch_nightly: true
   mirror_hardwares: [amdexperimental, amdproduction]
   agent_pool: mi325_1
   # grade: Blocking
   source_file_dependencies:
+  - vllm/engine/arg_utils.py
+  - vllm/config/model.py
   - vllm/model_executor
   - tests/model_executor
   - tests/entrypoints/openai/test_tensorizer_entrypoint.py
@@ -861,9 +919,10 @@ steps:
     - 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
 
 - label: Multi-Modal Accuracy Eval (Small Models) # 10min
+  timeout_in_minutes: 70
   mirror_hardwares: [amdexperimental, amdproduction]
   agent_pool: mi325_1
-  timeout_in_minutes: 15
+  # grade: Blocking
   working_dir: "/vllm-workspace/.buildkite/lm-eval-harness"
   source_file_dependencies:
   - vllm/multimodal/
@@ -934,6 +993,7 @@ steps:
 - label: Transformers Nightly Models Test
   mirror_hardwares: [amdexperimental]
   agent_pool: mi325_1
+  # grade: Blocking
   working_dir: "/vllm-workspace/"
   optional: true
   commands:
@@ -961,11 +1021,16 @@ steps:
   - vllm/model_executor/layers/fused_moe/flashinfer_cutlass_prepare_finalize.py
   - vllm/model_executor/layers/quantization/utils/flashinfer_utils.py
   - vllm/v1/attention/backends/flashinfer.py
+  - vllm/v1/attention/backends/mla/cutlass_mla.py
+  - vllm/v1/attention/backends/mla/flashinfer_mla.py
+  - vllm/platforms/cuda.py
+  - vllm/attention/selector.py
   commands:
     - nvidia-smi
     - python3 examples/offline_inference/basic/chat.py
     # Attention
     # num_heads2 broken by https://github.com/flashinfer-ai/flashinfer/issues/1353
+    - pytest -v -s tests/kernels/attention/test_attention_selector.py
     - pytest -v -s tests/kernels/attention/test_flashinfer.py -k 'not num_heads2'
     - pytest -v -s tests/kernels/attention/test_flashinfer_trtllm_attention.py
     - pytest -v -s tests/kernels/attention/test_cutlass_mla_decode.py
@@ -1002,7 +1067,33 @@ steps:
     - pytest -v -s tests/compile/test_silu_mul_quant_fusion.py
     # this runner has 2 GPUs available even though num_gpus=2 is not set
     - pytest -v -s tests/compile/test_fusion_all_reduce.py
+    # Limit to Inductor partition, no custom ops, and allreduce & attn fusion to reduce running time
+    # Wrap with quotes to escape yaml 
+    - "pytest -v -s tests/compile/test_fusions_e2e.py::test_tp2_attn_quant_allreduce_rmsnorm -k 'True and Llama-3.1 and -quant_fp8 and -rms_norm'"
+
+- label: Blackwell Fusion E2E Tests # 30 min
+  timeout_in_minutes: 40
+  working_dir: "/vllm-workspace/"
+  gpu: b200
+  optional: true
+  num_gpus: 2
+  source_file_dependencies:
+  - csrc/quantization/fp4/
+  - vllm/model_executor/layers/quantization/utils/flashinfer_utils.py
+  - vllm/v1/attention/backends/flashinfer.py
+  - vllm/compilation/
+  # can affect pattern matching
+  - vllm/model_executor/layers/layernorm.py
+  - vllm/model_executor/layers/activation.py
+  - vllm/model_executor/layers/quantization/input_quant_fp8.py
+  - tests/compile/test_fusions_e2e.py
+  - tests/compile/test_full_graph.py
+  commands:
+    - nvidia-smi
+    # Run all e2e fusion tests
     - pytest -v -s tests/compile/test_fusions_e2e.py
+    # test_fp8_kv_scale_compile requires FlashAttention (not supported on default L4/L40)
+    - pytest -v -s tests/compile/test_full_graph.py::test_fp8_kv_scale_compile
 
 - label: Blackwell GPT-OSS Eval
   timeout_in_minutes: 60
@@ -1253,6 +1344,7 @@ steps:
 - label: NixlConnector PD accuracy tests (Distributed) # 30min
   mirror_hardwares: [amdexperimental]
   agent_pool: mi325_4
+  # grade: Blocking
   timeout_in_minutes: 30
   working_dir: "/vllm-workspace/tests"
   num_gpus: 4
@@ -1267,6 +1359,9 @@ steps:
 ##### A100 test #####
 
 - label: Distributed Tests (A100) # optional
+  mirror_hardwares: [amdexperimental]
+  agent_pool: mi325_4
+  # grade: Blocking
   gpu: a100
   optional: true
   num_gpus: 4
@@ -1281,6 +1376,9 @@ steps:
   - pytest -v -s -x lora/test_mixtral.py
 
 - label: LM Eval Large Models # optional
+  mirror_hardwares: [amdexperimental, amdproduction]
+  agent_pool: mi325_4
+  # grade: Blocking
   gpu: a100
   optional: true
   num_gpus: 4
@@ -1292,8 +1390,27 @@ steps:
   - export VLLM_WORKER_MULTIPROC_METHOD=spawn
   - pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-large.txt --tp-size=4
 
+##### H100 test #####
+- label: LM Eval Large Models (H100) # optional
+  mirror_hardwares: [amdexperimental, amdproduction]
+  agent_pool: mi325_4
+  # grade: Blocking
+  gpu: h100
+  optional: true
+  num_gpus: 4
+  working_dir: "/vllm-workspace/.buildkite/lm-eval-harness"
+  source_file_dependencies:
+  - csrc/
+  - vllm/model_executor/layers/quantization
+  commands:
+    - export VLLM_USE_DEEP_GEMM=0  # We found Triton is faster than DeepGEMM for H100
+    - pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-large-hopper.txt --tp-size=4
+
 ##### H200 test #####
 - label: Distributed Tests (H200) # optional
+  mirror_hardwares: [amdexperimental]
+  agent_pool: mi325_2
+  # grade: Blocking
   gpu: h200
   optional: true
   working_dir: "/vllm-workspace/"
@@ -1305,6 +1422,7 @@ steps:
     - pytest -v -s tests/compile/test_fusions_e2e.py::test_tp2_attn_quant_allreduce_rmsnorm
     - pytest -v -s tests/distributed/test_context_parallel.py
     - CUDA_VISIBLE_DEVICES=1,2 VLLM_ALL2ALL_BACKEND=deepep_high_throughput VLLM_USE_DEEP_GEMM=1 VLLM_LOGGING_LEVEL=DEBUG python3 examples/offline_inference/data_parallel.py --model Qwen/Qwen1.5-MoE-A2.7B --tp-size=1  --dp-size=2 --max-model-len 2048
+    - pytest -v -s tests/v1/distributed/test_dbo.py
 
 ##### B200 test #####
 - label: Distributed Tests (B200) # optional
@@ -1315,6 +1433,7 @@ steps:
   commands:
     - pytest -v -s tests/distributed/test_context_parallel.py
     - pytest -v -s tests/distributed/test_nccl_symm_mem_allreduce.py
+    - pytest -v -s tests/v1/distributed/test_dbo.py
 
 ##### RL Integration Tests #####
 - label: Prime-RL Integration Test # 15min
@@ -1330,3 +1449,27 @@ steps:
   - .buildkite/scripts/run-prime-rl-test.sh
   commands:
     - bash .buildkite/scripts/run-prime-rl-test.sh
+
+- label: DeepSeek V2-Lite Accuracy
+  mirror_hardwares: [amdexperimental]
+  agent_pool: mi325_4
+  # grade: Blocking
+  timeout_in_minutes: 60
+  gpu: h100
+  optional: true
+  num_gpus: 4
+  working_dir: "/vllm-workspace"
+  commands:
+  - bash .buildkite/scripts/scheduled_integration_test/deepseek_v2_lite_ep_eplb.sh 0.25 200 8010
+
+- label: Qwen3-30B-A3B-FP8-block Accuracy
+  mirror_hardwares: [amdexperimental]
+  agent_pool: mi325_4
+  # grade: Blocking
+  timeout_in_minutes: 60
+  gpu: h100
+  optional: true
+  num_gpus: 4
+  working_dir: "/vllm-workspace"
+  commands:
+  - bash .buildkite/scripts/scheduled_integration_test/qwen30b_a3b_fp8_block_ep.sh 0.8 200 8020