@tduncan / Nvidia-Optimization-Stack-VLLM
Code Commits Issues Pulls Cloud Agent CI/CD
main
main
yaml 152 lines 4.93 KB
Raw
1 group: E2E Integration
2 depends_on:
3 - image-build
4 steps:
5 - label: ":nvidia: (H100) DeepSeek V2-Lite Sync EPLB Accuracy"
6 key: deepseek-v2-lite-sync-eplb-accuracy-4xh100
7 timeout_in_minutes: 25
8 device: h100
9 optional: true
10 num_devices: 4
11 working_dir: "/vllm-workspace"
12 commands:
13 - bash .buildkite/scripts/scheduled_integration_test/deepseek_v2_lite_ep_eplb.sh 0.25 200 8010
14 mirror:
15 amd:
16 label: ":amd: (MI355) DeepSeek V2-Lite Sync EPLB Accuracy"
17 dind: false
18 device: mi355_4
19 timeout_in_minutes: 40
20 env:
21 VLLM_ENGINE_READY_TIMEOUT_S: "1800"
22 depends_on:
23 - image-build-amd
24 source_file_dependencies:
25 - vllm/model_executor/models/
26 - vllm/model_executor/model_loader/
27 - vllm/distributed/eplb
28 - vllm/model_executor/layers/fused_moe/
29 - vllm/model_executor/layers/quantization/
30 - vllm/v1/attention/backends/
31 - vllm/v1/attention/backends/mla/
32 - vllm/v1/attention/selector.py
33 - .buildkite/scripts/scheduled_integration_test/
34 - vllm/_aiter_ops.py
35 - vllm/platforms/rocm.py
36
37 - label: ":nvidia: (H100) Qwen3-30B-A3B-FP8-block Sync EPLB Accuracy"
38 key: qwen3-30b-a3b-fp8-block-sync-eplb-accuracy-4xh100
39 timeout_in_minutes: 25
40 device: h100
41 optional: true
42 num_devices: 4
43 working_dir: "/vllm-workspace"
44 commands:
45 - bash .buildkite/scripts/scheduled_integration_test/qwen30b_a3b_fp8_block_ep_eplb.sh 0.8 200 8020
46 mirror:
47 amd:
48 label: ":amd: (MI355) Qwen3-30B-A3B-FP8-block Sync EPLB Accuracy TP2"
49 dind: false
50 device: mi355_4
51 timeout_in_minutes: 30
52 depends_on:
53 - image-build-amd
54 source_file_dependencies:
55 - vllm/model_executor/models/
56 - vllm/model_executor/model_loader/
57 - vllm/model_executor/layers/quantization/
58 - vllm/distributed/eplb
59 - vllm/model_executor/layers/fused_moe/
60 - vllm/v1/attention/backends/
61 - vllm/v1/attention/selector.py
62 - .buildkite/scripts/scheduled_integration_test/
63 - vllm/_aiter_ops.py
64 - vllm/platforms/rocm.py
65
66 - label: ":nvidia: (B200) Qwen3-30B-A3B-FP8-block Sync EPLB Accuracy"
67 key: qwen3-30b-a3b-fp8-block-sync-eplb-accuracy-2xb200
68 timeout_in_minutes: 20
69 device: b200-k8s
70 optional: true
71 num_devices: 2
72 working_dir: "/vllm-workspace"
73 commands:
74 - bash .buildkite/scripts/scheduled_integration_test/qwen30b_a3b_fp8_block_ep_eplb.sh 0.8 200 8020 2 1
75 mirror:
76 amd:
77 label: ":amd: (MI355) Qwen3-30B-A3B-FP8-block Sync EPLB Accuracy TP1"
78 dind: false
79 device: mi355_2
80 timeout_in_minutes: 25
81 depends_on:
82 - image-build-amd
83 source_file_dependencies:
84 - vllm/model_executor/models/
85 - vllm/model_executor/model_loader/
86 - vllm/model_executor/layers/quantization/
87 - vllm/model_executor/layers/fused_moe/
88 - vllm/distributed/eplb
89 - vllm/v1/attention/backends/
90 - vllm/v1/attention/selector.py
91 - .buildkite/scripts/scheduled_integration_test/
92 - vllm/_aiter_ops.py
93 - vllm/platforms/rocm.py
94
95 - label: ":nvidia: (H100) Qwen3-30B-A3B-FP8 DP4 Async EPLB Accuracy"
96 key: qwen3-30b-a3b-fp8-dp4-async-eplb-accuracy
97 timeout_in_minutes: 25
98 device: h100
99 optional: true
100 num_devices: 4
101 working_dir: "/vllm-workspace"
102 commands:
103 - bash .buildkite/scripts/scheduled_integration_test/qwen30b_a3b_fp8_dp4_async_eplb.sh 0.8 200 8050
104 mirror:
105 amd:
106 label: ":amd: (MI355) Qwen3-30B-A3B-FP8 DP4 Async EPLB Accuracy"
107 dind: false
108 device: mi355_4
109 timeout_in_minutes: 30
110 depends_on:
111 - image-build-amd
112 source_file_dependencies:
113 - vllm/model_executor/models/
114 - vllm/model_executor/model_loader/
115 - vllm/model_executor/layers/quantization/
116 - vllm/distributed/eplb
117 - vllm/model_executor/layers/fused_moe/
118 - vllm/v1/attention/backends/
119 - vllm/v1/attention/selector.py
120 - .buildkite/scripts/scheduled_integration_test/
121 - vllm/_aiter_ops.py
122 - vllm/platforms/rocm.py
123
124 - label: ":nvidia: (H100) DeepSeek V2-Lite Prefetch Offload Accuracy"
125 key: deepseek-v2-lite-prefetch-offload-accuracy-h100
126 timeout_in_minutes: 20
127 device: h100
128 optional: true
129 num_devices: 1
130 working_dir: "/vllm-workspace"
131 commands:
132 - bash .buildkite/scripts/scheduled_integration_test/deepseek_v2_lite_prefetch_offload.sh 0.25 200 8030
133 mirror:
134 amd:
135 label: ":amd: (MI355 DPX) DeepSeek V2-Lite Prefetch Offload Accuracy"
136 dind: false
137 device: mi355_dpx
138 working_dir: "/vllm-workspace"
139 timeout_in_minutes: 30
140 depends_on:
141 - image-build-amd
142 source_file_dependencies:
143 - vllm/model_executor/models/
144 - vllm/model_executor/model_loader/
145 - vllm/model_executor/layers/fused_moe/
146 - vllm/model_executor/layers/quantization/
147 - vllm/v1/attention/backends/
148 - vllm/v1/attention/backends/mla/
149 - vllm/v1/attention/selector.py
150 - .buildkite/scripts/scheduled_integration_test/
151 - vllm/_aiter_ops.py
152 - vllm/platforms/rocm.py