@tduncan / Nvidia-Optimization-Stack-VLLM
Code Commits Issues Pulls Cloud Agent CI/CD
main
main
yaml 167 lines 4.67 KB
Raw
1 group: LoRA Intel
2 depends_on:
3 - image-build-xpu
4 steps:
5 - label: LoRA Runtime + Utils
6 key: lora-runtime-utils
7 timeout_in_minutes: 45
8 device: intel_gpu
9 agent_tags:
10 label: production
11 gpu: 1+
12 mem: 24+
13 no_plugin: true
14 working_dir: "."
15 env:
16 REGISTRY: "public.ecr.aws/q9t5s3a7"
17 REPO: "vllm-ci-test-repo"
18 VLLM_TEST_DEVICE: "xpu"
19 source_file_dependencies:
20 - vllm/lora
21 - tests/lora
22 commands:
23 - >-
24 bash .buildkite/scripts/hardware_ci/run-intel-test.sh
25 'cd tests &&
26 export VLLM_WORKER_MULTIPROC_METHOD=spawn &&
27 pytest -v -s lora/test_layers.py &&
28 pytest -v -s lora/test_lora_checkpoints.py &&
29 pytest -v -s lora/test_lora_functions.py &&
30 pytest -v -s lora/test_lora_huggingface.py &&
31 pytest -v -s lora/test_lora_manager.py &&
32 pytest -v -s lora/test_lora_utils.py &&
33 pytest -v -s lora/test_peft_helper.py &&
34 pytest -v -s lora/test_resolver.py &&
35 pytest -v -s lora/test_utils.py &&
36 pytest -v -s lora/test_add_lora.py &&
37 pytest -v -s lora/test_worker.py'
38
39 - label: LoRA Fused/MoE Kernels
40 key: lora-fused-moe-kernels
41 timeout_in_minutes: 45
42 device: intel_gpu
43 agent_tags:
44 label: production
45 gpu: 1+
46 mem: 16+
47 no_plugin: true
48 working_dir: "."
49 env:
50 REGISTRY: "public.ecr.aws/q9t5s3a7"
51 REPO: "vllm-ci-test-repo"
52 VLLM_TEST_DEVICE: "xpu"
53 source_file_dependencies:
54 - vllm/lora
55 - tests/lora
56 commands:
57 - >-
58 bash .buildkite/scripts/hardware_ci/run-intel-test.sh
59 'cd tests &&
60 export VLLM_WORKER_MULTIPROC_METHOD=spawn &&
61 pytest -v -s lora/test_fused_moe_lora_kernel.py &&
62 pytest -v -s lora/test_moe_lora_align_sum.py --deselect="tests/lora/test_moe_lora_align_sum.py::test_moe_lora_align_block_size_mixed_base_and_lora[1]"'
63
64 - label: LoRA Punica Kernels
65 key: lora-punica-kernels
66 timeout_in_minutes: 45
67 device: intel_gpu
68 agent_tags:
69 label: production
70 gpu: 1+
71 mem: 16+
72 no_plugin: true
73 working_dir: "."
74 env:
75 REGISTRY: "public.ecr.aws/q9t5s3a7"
76 REPO: "vllm-ci-test-repo"
77 VLLM_TEST_DEVICE: "xpu"
78 source_file_dependencies:
79 - vllm/lora
80 - tests/lora
81 commands:
82 - >-
83 bash .buildkite/scripts/hardware_ci/run-intel-test.sh
84 'cd tests &&
85 export VLLM_WORKER_MULTIPROC_METHOD=spawn &&
86 set -o pipefail &&
87 pytest -v -s lora/test_punica_ops.py::test_kernels &&
88 pytest -v -s lora/test_punica_ops.py::test_kernels_hidden_size &&
89 pytest -v -s lora/test_punica_ops.py::test_add_lora_fused_moe_early_exit'
90
91 - label: LoRA Punica FP8/XPU Ops
92 key: lora-punica-fp8-xpu-ops
93 timeout_in_minutes: 60
94 device: intel_gpu
95 agent_tags:
96 label: production
97 gpu: 1+
98 mem: 16+
99 no_plugin: true
100 working_dir: "."
101 env:
102 REGISTRY: "public.ecr.aws/q9t5s3a7"
103 REPO: "vllm-ci-test-repo"
104 VLLM_TEST_DEVICE: "xpu"
105 source_file_dependencies:
106 - vllm/lora
107 - tests/lora
108 commands:
109 - >-
110 bash .buildkite/scripts/hardware_ci/run-intel-test.sh
111 'cd tests &&
112 export VLLM_WORKER_MULTIPROC_METHOD=spawn &&
113 pytest -v -s lora/test_punica_ops_fp8.py &&
114 pytest -v -s lora/test_punica_xpu_ops.py'
115
116 - label: LoRA Models
117 key: lora-models
118 timeout_in_minutes: 45
119 device: intel_gpu
120 agent_tags:
121 label: production
122 gpu: 2+
123 mem: 24+
124 no_plugin: true
125 working_dir: "."
126 env:
127 REGISTRY: "public.ecr.aws/q9t5s3a7"
128 REPO: "vllm-ci-test-repo"
129 VLLM_TEST_DEVICE: "xpu"
130 source_file_dependencies:
131 - vllm/lora
132 - tests/lora
133 commands:
134 - >-
135 bash .buildkite/scripts/hardware_ci/run-intel-test.sh
136 'cd tests &&
137 export VLLM_WORKER_MULTIPROC_METHOD=spawn &&
138 pytest -v -s lora/test_quant_model.py --deselect="tests/lora/test_quant_model.py::test_quant_model_lora[model0]" --deselect="tests/lora/test_quant_model.py::test_quant_model_lora[model1]" --deselect="tests/lora/test_quant_model.py::test_quant_model_tp_equality[model0]" &&
139 pytest -v -s lora/test_transformers_model.py &&
140 pytest -v -s lora/test_chatglm3_tp.py &&
141 pytest -v -s lora/test_llama_tp.py::test_llama_lora &&
142 pytest -s -v lora/test_minicpmv_tp.py'
143
144 - label: LoRA Multimodal
145 key: lora-multimodal
146 timeout_in_minutes: 45
147 device: intel_gpu
148 agent_tags:
149 label: production
150 gpu: 1+
151 mem: 24+
152 no_plugin: true
153 working_dir: "."
154 env:
155 REGISTRY: "public.ecr.aws/q9t5s3a7"
156 REPO: "vllm-ci-test-repo"
157 VLLM_TEST_DEVICE: "xpu"
158 source_file_dependencies:
159 - vllm/lora
160 - tests/lora
161 commands:
162 - >-
163 bash .buildkite/scripts/hardware_ci/run-intel-test.sh
164 'cd tests &&
165 export VLLM_WORKER_MULTIPROC_METHOD=spawn &&
166 pytest -v -s lora/test_default_mm_loras.py &&
167 pytest -v -s lora/test_whisper.py'