@tduncan / Nvidia-Optimization-Stack-VLLM
Code Commits Issues Pulls Cloud Agent CI/CD
main
main
yaml 78 lines 2.94 KB
Raw
1 group: Models - Distributed
2 depends_on:
3 - image-build
4 steps:
5 - label: ":nvidia: (L4) Distributed Models Shard %N"
6 key: distributed-model-tests-2-gpus
7 timeout_in_minutes: 35
8 working_dir: "/vllm-workspace/tests"
9 device: l4
10 num_devices: 2
11 source_file_dependencies:
12 - vllm/model_executor/model_loader/sharded_state_loader.py
13 - vllm/model_executor/models/
14 - tests/basic_correctness/
15 - tests/model_executor/model_loader/test_sharded_state_loader.py
16 - tests/models/
17 commands:
18 - |
19 case "$$BUILDKITE_PARALLEL_JOB" in
20 0)
21 TARGET_TEST_SUITE=L4 pytest basic_correctness/ -v -s -m 'distributed(num_gpus=2)'
22 CUDA_VISIBLE_DEVICES=0,1 pytest -v -s model_executor/model_loader/test_sharded_state_loader.py -m '(not slow_test)'
23 ;;
24 1)
25 pytest models/transformers/test_backend.py -v -s -m 'distributed(num_gpus=2)'
26 pytest models/language -v -s -m 'distributed(num_gpus=2)'
27 ;;
28 2)
29 pytest models/multimodal -v -s -m 'distributed(num_gpus=2)' --ignore models/multimodal/generation/test_whisper.py
30 VLLM_WORKER_MULTIPROC_METHOD=spawn pytest models/multimodal/generation/test_whisper.py -v -s -m 'distributed(num_gpus=2)'
31 ;;
32 *) exit 2 ;;
33 esac
34 parallelism: 3
35 mirror:
36 amd:
37 label: ":amd: (MI300) Distributed Models Shard %N"
38 dind: false
39 device: mi300_2
40 timeout_in_minutes: 80
41 depends_on:
42 - image-build-amd
43 source_file_dependencies:
44 - vllm/models/hy_v4/
45 - vllm/config/
46 - vllm/v1/worker/
47 - vllm/v1/attention/ops/
48 - vllm/v1/attention/backend.py
49 - tests/conftest.py
50 - tests/utils.py
51 - vllm/model_executor/model_loader/sharded_state_loader.py
52 - vllm/model_executor/models/
53 - vllm/model_executor/layers/
54 - vllm/v1/attention/backends/
55 - vllm/v1/attention/selector.py
56 - vllm/_aiter_ops.py
57 - vllm/platforms/rocm.py
58 - tests/basic_correctness/
59 - tests/model_executor/model_loader/test_sharded_state_loader.py
60 - tests/models/
61 commands:
62 - |
63 case "$$BUILDKITE_PARALLEL_JOB" in
64 0)
65 TARGET_TEST_SUITE=MI300 pytest basic_correctness/ -v -s -m 'distributed(num_gpus=2)' &&
66 HIP_VISIBLE_DEVICES=0,1 pytest -v -s model_executor/model_loader/test_sharded_state_loader.py -m '(not slow_test)'
67 ;;
68 1)
69 pytest models/transformers/test_backend.py -v -s -m 'distributed(num_gpus=2)' &&
70 pytest models/language -v -s -m 'distributed(num_gpus=2)' &&
71 VLLM_ROCM_USE_AITER=1 pytest -v -s models/test_hyv4_rocm.py -m 'distributed(num_gpus=2)'
72 ;;
73 2)
74 pytest models/multimodal -v -s -m 'distributed(num_gpus=2)' --ignore models/multimodal/generation/test_whisper.py &&
75 VLLM_WORKER_MULTIPROC_METHOD=spawn pytest models/multimodal/generation/test_whisper.py -v -s -m 'distributed(num_gpus=2)'
76 ;;
77 *) exit 2 ;;
78 esac