Mirroring changes in test-pipeline.yaml into test-amd.yaml #27242
Conversation
Signed-off-by: Alexei V. Ivanov <[email protected]>
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request mirrors changes from test-pipeline.yaml into test-amd.yaml. While many changes are valid, several NVIDIA-specific test configurations (for Blackwell and H200 GPUs) have been incorrectly copied into the AMD pipeline file. These steps will fail on AMD hardware and should be removed. There is also a CUDA-specific test dependency that should be removed.
| - label: Blackwell Test # 21 min | ||
| timeout_in_minutes: 30 |
There was a problem hiding this comment.
While this change updates the label and timeout, this entire 'Blackwell Test' step is configured for NVIDIA hardware (gpu: b200 on line 943) and is not suitable for the test-amd.yaml pipeline. This step should be removed entirely from this file.
| - label: Blackwell Fusion Tests # 30 min | ||
| timeout_in_minutes: 40 | ||
| working_dir: "/vllm-workspace/" | ||
| gpu: b200 | ||
| 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 | ||
| commands: | ||
| - nvidia-smi | ||
| - pytest -v -s tests/compile/test_fusion_attn.py | ||
| - 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 | ||
| - pytest -v -s tests/compile/test_fusions_e2e.py |
There was a problem hiding this comment.
This newly added 'Blackwell Fusion Tests' step is configured for NVIDIA Blackwell (B200) GPUs and includes the nvidia-smi command. This is incorrect for the test-amd.yaml pipeline, which is intended for AMD hardware. This entire step should be removed.
| - label: Distributed Tests (H200) # optional | ||
| gpu: h200 | ||
| optional: true | ||
| working_dir: "/vllm-workspace/" | ||
| num_gpus: 2 | ||
| commands: | ||
| - pytest -v -s tests/compile/test_async_tp.py | ||
| - pytest -v -s tests/compile/test_sequence_parallelism.py | ||
| - pytest -v -s tests/compile/test_fusion_all_reduce.py | ||
| - 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 |
There was a problem hiding this comment.
This test step is configured for NVIDIA H200 GPUs (gpu: h200) and is not suitable for the test-amd.yaml pipeline. While the changes mirror another file, this entire step, including the typo fix and newly added tests, should be removed from the AMD-specific configuration.
| source_file_dependencies: | ||
| - csrc/ | ||
| - tests/kernels/core | ||
| - tests/kernels/test_top_k_per_row.py |
There was a problem hiding this comment.
The test file tests/kernels/test_top_k_per_row.py is CUDA-specific, as indicated by @pytest.mark.skipif(not current_platform.is_cuda(), reason="This test requires CUDA") within the file. Including this as a source dependency in test-amd.yaml is incorrect. This dependency should be removed from the AMD pipeline.
| - 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) # 50min | ||
| mirror_hardwares: [amdexperimental] |
There was a problem hiding this comment.
why don't we use amd production to enable this in actual CI?
There was a problem hiding this comment.
This pipeline in its current state is meant to be for monitoring only and not gating for the PRs
|
|
||
| - label: Blackwell Test # 38 min | ||
| timeout_in_minutes: 60 | ||
| - label: Blackwell Test # 21 min |
There was a problem hiding this comment.
Do we still need to test Blackwell on this AMD specific pipeline?
There was a problem hiding this comment.
We include that test group for completeness reasons. For the moment it is just a comment and not an instruction for action in test-amd.yaml. But things may change in the future.
3 participants
Mirroring changes in test-pipeline.yaml into test-amd.yaml
Signed-off-by: Alexei V. Ivanov [email protected]