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, 11:00 UTC
[Test] Attempt all TPU V1 tests, even if some of them fail. (#17334)
Signed-off-by: Yarong Mu <ymu@google.com>
yarongmu-google committed
May 8, 2025 at 10:20 UTC
a83a0f92b56b71855dc38e8e3d9809619e58bcd1
1 file changed
+74
-30
.buildkite/scripts/hardware_ci/run-tpu-v1-test.sh
+74
-30
@@ -1,6 +1,6 @@
1
#!/bin/bash
2
3
-set -xue
3
+set -xu
4
5
# Build the docker image.
6
docker build -f docker/Dockerfile.tpu -t vllm-tpu .
@@ -24,36 +24,80 @@ docker run --privileged --net host --shm-size=16G -it \
24
&& export VLLM_XLA_CHECK_RECOMPILATION=1 \
25
&& echo HARDWARE \
26
&& tpu-info \
27
- && echo TEST_0 \
28
- && pytest -v -s /workspace/vllm/tests/v1/tpu/test_perf.py \
29
- && echo TEST_1 \
30
- && pytest -v -s /workspace/vllm/tests/tpu/test_compilation.py \
31
- && echo TEST_2 \
32
- && pytest -v -s /workspace/vllm/tests/v1/tpu/test_basic.py \
33
- && echo TEST_3 \
34
- && pytest -v -s /workspace/vllm/tests/entrypoints/llm/test_accuracy.py::test_lm_eval_accuracy_v1_engine \
35
- && echo TEST_4 \
36
- && pytest -s -v /workspace/vllm/tests/tpu/test_quantization_accuracy.py \
37
- && echo TEST_5 \
38
- && python3 /workspace/vllm/examples/offline_inference/tpu.py \
39
- && echo TEST_6 \
40
- && pytest -s -v /workspace/vllm/tests/v1/tpu/worker/test_tpu_model_runner.py \
41
- && echo TEST_7 \
42
- && pytest -s -v /workspace/vllm/tests/v1/tpu/test_sampler.py \
43
- && echo TEST_8 \
44
- && pytest -s -v /workspace/vllm/tests/v1/tpu/test_topk_topp_sampler.py \
45
- && echo TEST_9 \
46
- && pytest -s -v /workspace/vllm/tests/v1/tpu/test_multimodal.py \
47
- && echo TEST_10 \
48
- && pytest -s -v /workspace/vllm/tests/v1/tpu/test_pallas.py \
49
- && echo TEST_11 \
50
- && pytest -s -v /workspace/vllm/tests/v1/entrypoints/llm/test_struct_output_generate.py \
51
- && echo TEST_12 \
52
- && pytest -s -v /workspace/vllm/tests/tpu/test_moe_pallas.py" \
27
+ && { \
28
+ echo TEST_0: Running test_perf.py; \
29
+ pytest -s -v /workspace/vllm/tests/tpu/test_perf.py; \
30
+ echo TEST_0_EXIT_CODE: \$?; \
31
+ } & \
32
+ && { \
33
+ echo TEST_1: Running test_compilation.py; \
34
+ pytest -s -v /workspace/vllm/tests/tpu/test_compilation.py; \
35
+ echo TEST_1_EXIT_CODE: \$?; \
36
+ } & \
37
+ { \
38
+ echo TEST_2: Running test_basic.py; \
39
+ pytest -s -v /workspace/vllm/tests/v1/tpu/test_basic.py; \
40
+ echo TEST_2_EXIT_CODE: \$?; \
41
+ } & \
42
+ { \
43
+ echo TEST_3: Running test_accuracy.py::test_lm_eval_accuracy_v1_engine; \
44
+ pytest -s -v /workspace/vllm/tests/entrypoints/llm/test_accuracy.py::test_lm_eval_accuracy_v1_engine; \
45
+ echo TEST_3_EXIT_CODE: \$?; \
46
+ } & \
47
+ { \
48
+ echo TEST_4: Running test_quantization_accuracy.py; \
49
+ pytest -s -v /workspace/vllm/tests/tpu/test_quantization_accuracy.py; \
50
+ echo TEST_4_EXIT_CODE: \$?; \
51
+ } & \
52
+ { \
53
+ echo TEST_5: Running examples/offline_inference/tpu.py; \
54
+ python3 /workspace/vllm/examples/offline_inference/tpu.py; \
55
+ echo TEST_5_EXIT_CODE: \$?; \
56
+ } & \
57
+ { \
58
+ echo TEST_6: Running test_tpu_model_runner.py; \
59
+ pytest -s -v /workspace/vllm/tests/tpu/worker/test_tpu_model_runner.py; \
60
+ echo TEST_6_EXIT_CODE: \$?; \
61
+ } & \
62
+ && { \
63
+ echo TEST_7: Running test_sampler.py; \
64
+ pytest -s -v /workspace/vllm/tests/v1/tpu/test_sampler.py; \
65
+ echo TEST_7_EXIT_CODE: \$?; \
66
+ } & \
67
+ && { \
68
+ echo TEST_8: Running test_topk_topp_sampler.py; \
69
+ pytest -s -v /workspace/vllm/tests/v1/tpu/test_topk_topp_sampler.py; \
70
+ echo TEST_8_EXIT_CODE: \$?; \
71
+ } & \
72
+ && { \
73
+ echo TEST_9: Running test_multimodal.py; \
74
+ pytest -s -v /workspace/vllm/tests/v1/tpu/test_multimodal.py; \
75
+ echo TEST_9_EXIT_CODE: \$?; \
76
+ } & \
77
+ && { \
78
+ echo TEST_10: Running test_pallas.py; \
79
+ pytest -s -v /workspace/vllm/tests/v1/tpu/test_pallas.py; \
80
+ echo TEST_10_EXIT_CODE: \$?; \
81
+ } & \
82
+ && { \
83
+ echo TEST_11: Running test_struct_output_generate.py; \
84
+ pytest -s -v /workspace/vllm/tests/v1/entrypoints/llm/test_struct_output_generate.py; \
85
+ echo TEST_11_EXIT_CODE: \$?; \
86
+ } & \
87
+ && { \
88
+ echo TEST_12: Running test_moe_pallas.py; \
89
+ pytest -s -v /workspace/vllm/tests/tpu/test_moe_pallas.py; \
90
+ echo TEST_12_EXIT_CODE: \$?; \
91
+ } & \
92
# Disable the TPU LoRA tests until the feature is activated
54
- # && echo TEST_13 \
55
- # && pytest -s -v /workspace/vllm/tests/tpu/lora/" \
56
-
93
+ # && { \
94
+ # echo TEST_13: Running test_moe_pallas.py; \
95
+ # pytest -s -v /workspace/vllm/tests/tpu/lora/; \
96
+ # echo TEST_13_EXIT_CODE: \$?; \
97
+ # } & \
98
+ wait \
99
+ && echo 'All tests have attempted to run. Check logs for individual test statuses and exit codes.' \
100
+"
101
102
# TODO: This test fails because it uses RANDOM_SEED sampling
103
# && VLLM_USE_V1=1 pytest -v -s /workspace/vllm/tests/tpu/test_custom_dispatcher.py \