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 4, 06:15 UTC
[CI] Route part of B200 jobs to b200-k8s (#41453)
Signed-off-by: khluu <khluu000@gmail.com> Co-authored-by: OpenAI Codex <noreply@openai.com>
Kevin H. Luu committed
May 5, 2026 at 19:00 UTC
f6537612521df1156d8ac13f524e427aca908322
10 files changed
+15
-15
.buildkite/test_areas/attention.yaml
+1
-1
@@ -17,7 +17,7 @@ steps:
17
- label: V1 attention (B200)
18
key: v1-attention-b200
19
timeout_in_minutes: 30
20
- device: b200
20
+ device: b200-k8s
21
source_file_dependencies:
22
- vllm/config/attention.py
23
- vllm/model_executor/layers/attention
.buildkite/test_areas/benchmarks.yaml
+1
-1
@@ -14,7 +14,7 @@ steps:
14
15
- label: Attention Benchmarks Smoke Test (B200)
16
key: attention-benchmarks-smoke-test-b200
17
- device: b200
17
+ device: b200-k8s
18
num_gpus: 2
19
optional: true
20
working_dir: "/vllm-workspace/"
.buildkite/test_areas/compile.yaml
+4
-4
@@ -43,7 +43,7 @@ steps:
43
key: asynctp-correctness-tests-b200
44
timeout_in_minutes: 50
45
working_dir: "/vllm-workspace/"
46
- device: b200
46
+ device: b200-k8s
47
optional: true
48
num_devices: 2
49
commands:
@@ -68,7 +68,7 @@ steps:
68
key: fusion-and-compile-unit-tests-2xb200
69
timeout_in_minutes: 20
70
working_dir: "/vllm-workspace/"
71
- device: b200
71
+ device: b200-k8s
72
source_file_dependencies:
73
- csrc/quantization/fp4/
74
- vllm/model_executor/layers/quantization/
@@ -137,7 +137,7 @@ steps:
137
key: fusion-e2e-config-sweep-b200
138
timeout_in_minutes: 30
139
working_dir: "/vllm-workspace/"
140
- device: b200
140
+ device: b200-k8s
141
num_devices: 1
142
optional: true
143
commands:
@@ -209,7 +209,7 @@ steps:
209
key: fusion-e2e-tp2-b200
210
timeout_in_minutes: 20
211
working_dir: "/vllm-workspace/"
212
- device: b200
212
+ device: b200-k8s
213
num_devices: 2
214
source_file_dependencies:
215
- csrc/quantization/
.buildkite/test_areas/distributed.yaml
+1
-1
@@ -212,7 +212,7 @@ steps:
212
213
- label: Distributed Tests (2 GPUs)(B200)
214
key: distributed-tests-2-gpus-b200
215
- device: b200
215
+ device: b200-k8s
216
optional: true
217
working_dir: "/vllm-workspace/"
218
num_devices: 2
.buildkite/test_areas/e2e_integration.yaml
+1
-1
@@ -25,7 +25,7 @@ steps:
25
- label: Qwen3-30B-A3B-FP8-block Accuracy (B200)
26
key: qwen3-30b-a3b-fp8-block-accuracy-b200
27
timeout_in_minutes: 60
28
- device: b200
28
+ device: b200-k8s
29
optional: true
30
num_devices: 2
31
working_dir: "/vllm-workspace"
.buildkite/test_areas/kernels.yaml
+2
-2
@@ -125,7 +125,7 @@ steps:
125
key: kernels-b200
126
timeout_in_minutes: 30
127
working_dir: "/vllm-workspace/"
128
- device: b200
128
+ device: b200-k8s
129
# optional: true
130
source_file_dependencies:
131
- csrc/quantization/fp4/
@@ -212,7 +212,7 @@ steps:
212
- label: Kernels Fp4 MoE Test (B200)
213
key: kernels-fp4-moe-test-b200
214
timeout_in_minutes: 60
215
- device: b200
215
+ device: b200-k8s
216
num_devices: 1
217
optional: true
218
commands:
.buildkite/test_areas/lm_eval.yaml
+2
-2
@@ -51,7 +51,7 @@ steps:
51
- label: LM Eval Qwen3.5 Models (B200)
52
key: lm-eval-qwen3-5-models-b200
53
timeout_in_minutes: 120
54
- device: b200
54
+ device: b200-k8s
55
optional: true
56
num_devices: 2
57
source_file_dependencies:
@@ -84,7 +84,7 @@ steps:
84
85
- label: MoE Refactor Integration Test (B200 - TEMPORARY)
86
key: moe-refactor-integration-test-b200-temporary
87
- device: b200
87
+ device: b200-k8s
88
optional: true
89
num_devices: 2
90
commands:
.buildkite/test_areas/misc.yaml
+1
-1
@@ -224,7 +224,7 @@ steps:
224
- label: Batch Invariance (B200)
225
key: batch-invariance-b200
226
timeout_in_minutes: 30
227
- device: b200
227
+ device: b200-k8s
228
source_file_dependencies:
229
- vllm/v1/attention
230
- vllm/model_executor/layers
.buildkite/test_areas/quantization.yaml
+1
-1
@@ -25,7 +25,7 @@ steps:
25
key: quantized-moe-test-b200
26
timeout_in_minutes: 60
27
working_dir: "/vllm-workspace/"
28
- device: b200
28
+ device: b200-k8s
29
source_file_dependencies:
30
- tests/quantization/test_blackwell_moe.py
31
- vllm/model_executor/models/deepseek_v2.py
.buildkite/test_areas/spec_decode.yaml
+1
-1
@@ -75,7 +75,7 @@ steps:
75
- label: Spec Decode Draft Model Nightly B200
76
key: spec-decode-draft-model-nightly-b200
77
timeout_in_minutes: 30
78
- device: b200
78
+ device: b200-k8s
79
optional: true
80
source_file_dependencies:
81
- vllm/v1/spec_decode/