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, 02:30 UTC
[CI] Narrow pytorch.yaml compile job source dependencies (#42057)
Signed-off-by: khluu <khluu000@gmail.com>
Kevin H. Luu committed
May 8, 2026 at 14:43 UTC
f47ccc8b1c211e2639cee07baa2825f262577b14
1 file changed
+120
-5
.buildkite/test_areas/pytorch.yaml
+120
-5
@@ -6,7 +6,30 @@ steps:
6
key: pytorch-compilation-unit-tests
7
timeout_in_minutes: 10
8
source_file_dependencies:
9
- - vllm/
9
+ - vllm/__init__.py
10
+ - vllm/_aiter_ops.py
11
+ - vllm/_custom_ops.py
12
+ - vllm/compilation/
13
+ - vllm/config/
14
+ - vllm/distributed/
15
+ - vllm/engine/
16
+ - vllm/env_override.py
17
+ - vllm/envs.py
18
+ - vllm/forward_context.py
19
+ - vllm/inputs/
20
+ - vllm/ir/
21
+ - vllm/kernels/
22
+ - vllm/logger.py
23
+ - vllm/model_executor/
24
+ - vllm/multimodal/
25
+ - vllm/platforms/
26
+ - vllm/plugins/
27
+ - vllm/sampling_params.py
28
+ - vllm/sequence.py
29
+ - vllm/transformers_utils/
30
+ - vllm/triton_utils/
31
+ - vllm/utils/
32
+ - vllm/v1/
33
- tests/compile
34
commands:
35
# Run unit tests defined directly under compile/,
@@ -24,7 +47,30 @@ steps:
47
device: h100
48
num_devices: 1
49
source_file_dependencies:
27
- - vllm/
50
+ - vllm/__init__.py
51
+ - vllm/_aiter_ops.py
52
+ - vllm/_custom_ops.py
53
+ - vllm/compilation/
54
+ - vllm/config/
55
+ - vllm/distributed/
56
+ - vllm/engine/
57
+ - vllm/env_override.py
58
+ - vllm/envs.py
59
+ - vllm/forward_context.py
60
+ - vllm/inputs/
61
+ - vllm/ir/
62
+ - vllm/kernels/
63
+ - vllm/logger.py
64
+ - vllm/model_executor/
65
+ - vllm/multimodal/
66
+ - vllm/platforms/
67
+ - vllm/plugins/
68
+ - vllm/sampling_params.py
69
+ - vllm/sequence.py
70
+ - vllm/transformers_utils/
71
+ - vllm/triton_utils/
72
+ - vllm/utils/
73
+ - vllm/v1/
74
- tests/compile/h100/
75
commands:
76
- "find compile/h100/ -name 'test_*.py' -print0 | xargs -0 -n1 -I{} pytest -s -v '{}'"
@@ -33,7 +79,30 @@ steps:
79
key: pytorch-compilation-passes-unit-tests
80
timeout_in_minutes: 20
81
source_file_dependencies:
36
- - vllm/
82
+ - vllm/__init__.py
83
+ - vllm/_aiter_ops.py
84
+ - vllm/_custom_ops.py
85
+ - vllm/compilation/
86
+ - vllm/config/
87
+ - vllm/distributed/
88
+ - vllm/engine/
89
+ - vllm/env_override.py
90
+ - vllm/envs.py
91
+ - vllm/forward_context.py
92
+ - vllm/inputs/
93
+ - vllm/ir/
94
+ - vllm/kernels/
95
+ - vllm/logger.py
96
+ - vllm/model_executor/
97
+ - vllm/multimodal/
98
+ - vllm/platforms/
99
+ - vllm/plugins/
100
+ - vllm/sampling_params.py
101
+ - vllm/sequence.py
102
+ - vllm/transformers_utils/
103
+ - vllm/triton_utils/
104
+ - vllm/utils/
105
+ - vllm/v1/
106
- tests/compile/passes
107
commands:
108
- pytest -s -v compile/passes --ignore compile/passes/distributed
@@ -42,7 +111,30 @@ steps:
111
key: pytorch-fullgraph-smoke-test
112
timeout_in_minutes: 35
113
source_file_dependencies:
45
- - vllm/
114
+ - vllm/__init__.py
115
+ - vllm/_aiter_ops.py
116
+ - vllm/_custom_ops.py
117
+ - vllm/compilation/
118
+ - vllm/config/
119
+ - vllm/distributed/
120
+ - vllm/engine/
121
+ - vllm/env_override.py
122
+ - vllm/envs.py
123
+ - vllm/forward_context.py
124
+ - vllm/inputs/
125
+ - vllm/ir/
126
+ - vllm/kernels/
127
+ - vllm/logger.py
128
+ - vllm/model_executor/
129
+ - vllm/multimodal/
130
+ - vllm/platforms/
131
+ - vllm/plugins/
132
+ - vllm/sampling_params.py
133
+ - vllm/sequence.py
134
+ - vllm/transformers_utils/
135
+ - vllm/triton_utils/
136
+ - vllm/utils/
137
+ - vllm/v1/
138
- tests/compile
139
commands:
140
# Run smoke tests under fullgraph directory, except test_full_graph.py
@@ -56,7 +148,30 @@ steps:
148
timeout_in_minutes: 30
149
device: h200_18gb
150
source_file_dependencies:
59
- - vllm/
151
+ - vllm/__init__.py
152
+ - vllm/_aiter_ops.py
153
+ - vllm/_custom_ops.py
154
+ - vllm/compilation/
155
+ - vllm/config/
156
+ - vllm/distributed/
157
+ - vllm/engine/
158
+ - vllm/env_override.py
159
+ - vllm/envs.py
160
+ - vllm/forward_context.py
161
+ - vllm/inputs/
162
+ - vllm/ir/
163
+ - vllm/kernels/
164
+ - vllm/logger.py
165
+ - vllm/model_executor/
166
+ - vllm/multimodal/
167
+ - vllm/platforms/
168
+ - vllm/plugins/
169
+ - vllm/sampling_params.py
170
+ - vllm/sequence.py
171
+ - vllm/transformers_utils/
172
+ - vllm/triton_utils/
173
+ - vllm/utils/
174
+ - vllm/v1/
175
- tests/compile
176
commands:
177
# fp8 kv scales not supported on sm89, tested on Blackwell instead