@tduncan / Nvidia-Optimization-Stack-VLLM
Code Commits Issues Pulls Cloud Agent CI/CD
main
main
yaml 230 lines 8.69 KB
Raw
1 group: CPU
2 depends_on: []
3 steps:
4 - label: CPU-Kernel Tests Shard %N
5 key: cpu-kernel-tests
6 depends_on: []
7 device: intel_cpu
8 no_plugin: true
9 source_file_dependencies:
10 - csrc/cpu/
11 - cmake/cpu_extension.cmake
12 - CMakeLists.txt
13 - vllm/_custom_ops.py
14 - tests/kernels/attention/test_cpu_attn.py
15 - tests/v1/attention/test_group_head_counts.py
16 - tests/kernels/attention/test_amx_mla.py
17 - vllm/v1/attention/backends/mla/amx_mla.py
18 - vllm/model_executor/layers/attention/mla_attention.py
19 - tests/kernels/test_deepseek_v4_cpu_kernels.py
20 - tests/kernels/moe/test_cpu_fused_moe.py
21 - tests/kernels/moe/test_cpu_quant_fused_moe.py
22 - tests/kernels/moe/test_zen_cpu_int8_moe.py
23 - vllm/model_executor/layers/fused_moe/experts/cpu_moe.py
24 - tests/kernels/test_onednn.py
25 - tests/kernels/test_awq_int4_to_int8.py
26 - tests/kernels/quantization/test_cpu_fp8_scaled_mm.py
27 - tests/kernels/quantization/test_zen_da8w4.py
28 - vllm/model_executor/kernels/linear/mixed_precision/zentorch.py
29 - vllm/model_executor/layers/fused_moe/experts/zentorch_moe.py
30 - tests/kernels/mamba/cpu/test_cpu_gdn_ops.py
31 - tests/kernels/mamba/test_cpu_short_conv.py
32 - tests/kernels/mamba/test_causal_conv1d.py
33 - tests/kernels/mamba/test_mamba_ssm.py
34 - vllm/v1/sample/ops/topk_topp_triton.py
35 - vllm/v1/sample/ops/topk_topp_sampler.py
36 - tests/v1/sample/test_topk_topp_sampler.py
37 commands:
38 - |
39 bash .buildkite/scripts/hardware_ci/run-cpu-test.sh 30m "
40 pytest -x -v -s --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB \
41 tests/kernels/attention/test_cpu_attn.py \
42 tests/v1/attention/test_group_head_counts.py \
43 tests/kernels/attention/test_amx_mla.py \
44 tests/kernels/test_deepseek_v4_cpu_kernels.py \
45 tests/kernels/moe/test_cpu_fused_moe.py \
46 tests/kernels/moe/test_cpu_quant_fused_moe.py \
47 tests/kernels/moe/test_zen_cpu_int8_moe.py \
48 tests/kernels/mamba/test_cpu_short_conv.py \
49 tests/kernels/test_onednn.py \
50 tests/kernels/test_awq_int4_to_int8.py \
51 tests/kernels/quantization/test_cpu_fp8_scaled_mm.py \
52 tests/kernels/quantization/test_zen_da8w4.py \
53 tests/kernels/mamba/cpu/test_cpu_gdn_ops.py \
54 tests/kernels/mamba/test_causal_conv1d.py \
55 tests/kernels/mamba/test_mamba_ssm.py \
56 tests/v1/sample/test_topk_topp_sampler.py::TestTritonTopkTopp"
57 parallelism: 2
58
59 # Speculative decoding under a CPU platform. The unit tests already pass on
60 # CPU but no CPU job runs them; the e2e tests gate the engine path itself,
61 # which unit tests cannot reach.
62 - label: CPU-Spec Decode Tests
63 key: cpu-spec-decode-tests
64 depends_on: []
65 soft_fail: true
66 device: intel_cpu
67 no_plugin: true
68 source_file_dependencies:
69 - vllm/v1/spec_decode/
70 - vllm/v1/worker/cpu/
71 - vllm/v1/worker/cpu_model_runner.py
72 - vllm/v1/worker/gpu/spec_decode/
73 - vllm/v1/attention/backends/cpu_attn.py
74 - vllm/config/speculative.py
75 - vllm/model_executor/models/qwen3_5.py
76 - vllm/model_executor/models/qwen3_5_mtp.py
77 - vllm/model_executor/models/llama_eagle3.py
78 - vllm/model_executor/layers/mamba/gdn/qwen_gdn_linear_attn.py
79 - vllm/model_executor/layers/mamba/mamba_mixer2.py
80 - vllm/model_executor/layers/mamba/ops/cpu/
81 - tests/v1/e2e/test_cpu_spec_decode.py
82 - tests/v1/spec_decode/
83 - tests/v1/worker/
84 commands:
85 - |
86 bash .buildkite/scripts/hardware_ci/run-cpu-test.sh 30m "
87 pytest -x -v -s \
88 tests/v1/worker/test_gpu_autoregressive_speculator.py \
89 tests/v1/worker/test_gpu_extract_hidden_states_speculator.py \
90 tests/v1/worker/test_gpu_batch_ordering.py \
91 tests/v1/worker/test_gpu_rejection_sampler_chunking.py \
92 tests/v1/worker/test_eagle3_aux_hidden_states_pp.py \
93 tests/v1/worker/test_spec_decode_embed_sharing_pp.py \
94 tests/v1/spec_decode/test_adaptive_verification.py \
95 tests/v1/spec_decode/test_dflash2.py \
96 tests/v1/spec_decode/test_dflash_causality.py \
97 tests/v1/spec_decode/test_eagle_draft_attn_metadata.py \
98 tests/v1/spec_decode/test_draft_attention_backend_override.py \
99 tests/v1/spec_decode/test_draft_moe_backend_override.py \
100 tests/v1/spec_decode/test_ngram.py \
101 tests/v1/spec_decode/test_llm_base_proposer_sampling.py
102 unset VLLM_CPU_CI_ENV
103 pytest -x -v -s tests/v1/e2e/test_cpu_spec_decode.py"
104
105 # Note: SDE can't be downloaded from CI host because of AWS WAF
106 # - label: CPU-Compatibility Tests
107 # depends_on: []
108 # device: intel_cpu
109 # no_plugin: true
110 # source_file_dependencies:
111 # - cmake/cpu_extension.cmake
112 # - setup.py
113 # - vllm/platforms/cpu.py
114 # commands:
115 # - |
116 # bash .buildkite/scripts/hardware_ci/run-cpu-test.sh 20m "
117 # bash .buildkite/scripts/hardware_ci/run-cpu-compatibility-test.sh"
118
119 - label: CPU-Language Generation and Pooling Model Tests Shard %N
120 key: cpu-language-generation-and-pooling-model-tests
121 depends_on: []
122 device: intel_cpu
123 no_plugin: true
124 source_file_dependencies:
125 - csrc/cpu/
126 - vllm/
127 - tests/models/language/generation/
128 - tests/models/language/pooling/
129 - tests/v1/e2e/test_cpu_linear_attn_chunked_prefix.py
130 commands:
131 - |
132 bash .buildkite/scripts/hardware_ci/run-cpu-test.sh 25m "
133 pytest -x -v -s tests/models/language/generation tests/models/language/pooling tests/v1/e2e/test_cpu_linear_attn_chunked_prefix.py -m cpu_model --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB"
134 parallelism: 3
135
136 - label: CPU-Quantization Model Tests
137 key: cpu-quantization-model-tests
138 depends_on: []
139 device: intel_cpu
140 no_plugin: true
141 source_file_dependencies:
142 - csrc/cpu/
143 - vllm/model_executor/layers/quantization/auto_gptq.py
144 - vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_int8.py
145 - vllm/model_executor/kernels/linear/mixed_precision/cpu.py
146 - vllm/model_executor/kernels/linear/scaled_mm/cpu.py
147 - vllm/model_executor/layers/fused_moe/experts/cpu_moe.py
148 - tests/quantization/test_compressed_tensors.py
149 - tests/quantization/test_cpu_wna16.py
150 - tests/quantization/test_cpu_w8a8.py
151 commands:
152 - |
153 bash .buildkite/scripts/hardware_ci/run-cpu-test.sh 45m "
154 pytest -x -v -s tests/quantization/test_compressed_tensors.py::test_compressed_tensors_w8a8_logprobs
155 pytest -x -v -s tests/quantization/test_cpu_wna16.py
156 pytest -x -v -s tests/quantization/test_cpu_w8a8.py"
157
158 - label: CPU-Distributed Tests (PP+TP)
159 key: cpu-distributed-tests-pp-tp
160 depends_on: []
161 device: intel_cpu
162 no_plugin: true
163 source_file_dependencies: &cpu_distributed_deps
164 - csrc/cpu/shm.cpp
165 - vllm/v1/worker/cpu_worker.py
166 - vllm/v1/worker/gpu_worker.py
167 - vllm/v1/worker/cpu_model_runner.py
168 - vllm/v1/worker/gpu_model_runner.py
169 - vllm/platforms/cpu.py
170 - vllm/distributed/parallel_state.py
171 - vllm/distributed/device_communicators/cpu_communicator.py
172 - .buildkite/scripts/hardware_ci/run-cpu-distributed-smoke-test.sh
173 commands:
174 - |
175 bash .buildkite/scripts/hardware_ci/run-cpu-test.sh 10m "
176 bash .buildkite/scripts/hardware_ci/run-cpu-distributed-smoke-test.sh tp_pp"
177
178 - label: CPU-Distributed Tests (DP+TP)
179 key: cpu-distributed-tests-dp-tp
180 depends_on: []
181 device: intel_cpu
182 no_plugin: true
183 source_file_dependencies: *cpu_distributed_deps
184 commands:
185 - |
186 bash .buildkite/scripts/hardware_ci/run-cpu-test.sh 10m "
187 bash .buildkite/scripts/hardware_ci/run-cpu-distributed-smoke-test.sh dp_tp"
188
189 - label: CPU-Multi-Modal Model Tests %N
190 key: cpu-multi-modal-model-tests-n
191 depends_on: []
192 device: intel_cpu
193 no_plugin: true
194 source_file_dependencies:
195 # - vllm/
196 - vllm/model_executor/layers/rotary_embedding
197 - tests/models/multimodal/generation/
198 commands:
199 - |
200 bash .buildkite/scripts/hardware_ci/run-cpu-test.sh 45m "
201 pytest -x -v -s tests/models/multimodal/generation --ignore=tests/models/multimodal/generation/test_pixtral.py --ignore=tests/models/multimodal/generation/test_qwen2_5_vl.py -m cpu_model --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB"
202 parallelism: 4
203
204 - label: CPU-Qwen2.5-VL Multimodal Tests
205 key: cpu-qwen2-5-vl-multimodal-tests
206 depends_on: []
207 device: intel_cpu
208 no_plugin: true
209 source_file_dependencies:
210 # - vllm/
211 - vllm/model_executor/layers/rotary_embedding
212 - tests/models/multimodal/generation/
213 commands:
214 - |
215 bash .buildkite/scripts/hardware_ci/run-cpu-test.sh 40m "
216 VLLM_CI_ENV=0 pytest -x -v -s tests/models/multimodal/generation/test_qwen2_5_vl.py"
217
218 - label: "Arm CPU Test Shard %N"
219 key: arm-cpu-test
220 depends_on:
221 - cpu-arm64-image-build
222 soft_fail: false
223 device: arm_cpu
224 no_plugin: true
225 commands:
226 - >-
227 bash .buildkite/scripts/hardware_ci/run-cpu-test-arm.sh
228 "$$BUILDKITE_PARALLEL_JOB"
229 "$REGISTRY/$REPO:$BUILDKITE_COMMIT-arm64-cpu"
230 parallelism: 3