A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 22:15 UTC
[perf-benchmark] Fix ECR path for premerge benchmark (#13512)
Signed-off-by: <> Co-authored-by: EC2 Default User <ec2-user@ip-172-31-20-117.us-west-2.compute.internal>
Kevin H. Luu committed
Feb 18, 2025 at 23:56 UTC
3b05cd4555593d966bf4159040105f79e1f4eb77
1 file changed
+93
-7
.buildkite/nightly-benchmarks/benchmark-pipeline.yaml
+93
-7
@@ -10,18 +10,24 @@ steps:
10
- image: badouralix/curl-jq
11
command:
12
- sh .buildkite/nightly-benchmarks/scripts/wait-for-image.sh
13
-
13
+ - label: "Cleanup H100"
14
+ agents:
15
+ queue: H100
16
+ depends_on: ~
17
+ command: docker system prune -a --volumes --force
18
+
19
- label: "A100"
20
# skip: "use this flag to conditionally skip the benchmark step, useful for PR testing"
21
agents:
22
queue: A100
23
depends_on: wait-for-container-image
24
+ if: build.branch == "main"
25
plugins:
26
- kubernetes:
27
podSpec:
28
priorityClassName: perf-benchmark
29
containers:
24
- - image: public.ecr.aws/q9t5s3a7/${BUILDKITE_BRANCH:-main} == "main" && "vllm-ci-postmerge-repo" || "vllm-ci-test-repo"}:$BUILDKITE_COMMIT
30
+ - image: public.ecr.aws/q9t5s3a7/vllm-ci-postmerge-repo:$BUILDKITE_COMMIT
31
command:
32
- bash .buildkite/nightly-benchmarks/scripts/run-performance-benchmarks.sh
33
resources:
@@ -50,9 +56,10 @@ steps:
56
agents:
57
queue: H200
58
depends_on: wait-for-container-image
59
+ if: build.branch == "main"
60
plugins:
61
- docker#v5.12.0:
55
- image: public.ecr.aws/q9t5s3a7/${BUILDKITE_BRANCH:-main} == "main" && "vllm-ci-postmerge-repo" || "vllm-ci-test-repo"}:$BUILDKITE_COMMIT
62
+ image: public.ecr.aws/q9t5s3a7/vllm-ci-postmerge-repo:$BUILDKITE_COMMIT
63
command:
64
- bash
65
- .buildkite/nightly-benchmarks/scripts/run-performance-benchmarks.sh
@@ -70,20 +77,99 @@ steps:
77
#key: block-h100
78
#depends_on: ~
79
73
- - label: "Cleanup H100"
80
+ - label: "H100"
81
+ # skip: "use this flag to conditionally skip the benchmark step, useful for PR testing"
82
agents:
83
queue: H100
76
- depends_on: ~
77
- command: docker system prune -a --volumes --force
84
+ depends_on: wait-for-container-image
85
+ if: build.branch == "main"
86
+ plugins:
87
+ - docker#v5.12.0:
88
+ image: public.ecr.aws/q9t5s3a7/vllm-ci-postmerge-repo:$BUILDKITE_COMMIT
89
+ command:
90
+ - bash
91
+ - .buildkite/nightly-benchmarks/scripts/run-performance-benchmarks.sh
92
+ mount-buildkite-agent: true
93
+ propagate-environment: true
94
+ ipc: host
95
+ gpus: all # see CUDA_VISIBLE_DEVICES for actual GPUs used
96
+ volumes:
97
+ - /data/benchmark-hf-cache:/root/.cache/huggingface
98
+ environment:
99
+ - VLLM_USAGE_SOURCE
100
+ - HF_TOKEN
101
+
102
+ # Premerge benchmark
103
+ - label: "A100"
104
+ # skip: "use this flag to conditionally skip the benchmark step, useful for PR testing"
105
+ agents:
106
+ queue: A100
107
+ depends_on: wait-for-container-image
108
+ if: build.branch != "main"
109
+ plugins:
110
+ - kubernetes:
111
+ podSpec:
112
+ priorityClassName: perf-benchmark
113
+ containers:
114
+ - image: public.ecr.aws/q9t5s3a7/vllm-ci-test-repo:$BUILDKITE_COMMIT
115
+ command:
116
+ - bash .buildkite/nightly-benchmarks/scripts/run-performance-benchmarks.sh
117
+ resources:
118
+ limits:
119
+ nvidia.com/gpu: 8
120
+ volumeMounts:
121
+ - name: devshm
122
+ mountPath: /dev/shm
123
+ env:
124
+ - name: VLLM_USAGE_SOURCE
125
+ value: ci-test
126
+ - name: HF_TOKEN
127
+ valueFrom:
128
+ secretKeyRef:
129
+ name: hf-token-secret
130
+ key: token
131
+ nodeSelector:
132
+ nvidia.com/gpu.product: NVIDIA-A100-SXM4-80GB
133
+ volumes:
134
+ - name: devshm
135
+ emptyDir:
136
+ medium: Memory
137
+
138
+ - label: "H200"
139
+ # skip: "use this flag to conditionally skip the benchmark step, useful for PR testing"
140
+ agents:
141
+ queue: H200
142
+ depends_on: wait-for-container-image
143
+ if: build.branch != "main"
144
+ plugins:
145
+ - docker#v5.12.0:
146
+ image: public.ecr.aws/q9t5s3a7/vllm-ci-test-repo:$BUILDKITE_COMMIT
147
+ command:
148
+ - bash
149
+ - .buildkite/nightly-benchmarks/scripts/run-performance-benchmarks.sh
150
+ mount-buildkite-agent: true
151
+ propagate-environment: true
152
+ ipc: host
153
+ gpus: 4,5,6,7
154
+ volumes:
155
+ - /data/benchmark-hf-cache:/root/.cache/huggingface
156
+ environment:
157
+ - VLLM_USAGE_SOURCE
158
+ - HF_TOKEN
159
+
160
+ #- block: "Run H100 Benchmark"
161
+ #key: block-h100
162
+ #depends_on: ~
163
164
- label: "H100"
165
# skip: "use this flag to conditionally skip the benchmark step, useful for PR testing"
166
agents:
167
queue: H100
168
depends_on: wait-for-container-image
169
+ if: build.branch != "main"
170
plugins:
171
- docker#v5.12.0:
86
- image: public.ecr.aws/q9t5s3a7/${BUILDKITE_BRANCH:-main} == "main" && "vllm-ci-postmerge-repo" || "vllm-ci-test-repo"}:$BUILDKITE_COMMIT
172
+ image: public.ecr.aws/q9t5s3a7/vllm-ci-test-repo:$BUILDKITE_COMMIT
173
command:
174
- bash
175
- .buildkite/nightly-benchmarks/scripts/run-performance-benchmarks.sh