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 11, 07:30 UTC
[ci] Sync test areas yaml file with test-pipeline (#30862)
Signed-off-by: Kevin H. Luu <khluu000@gmail.com>
Kevin H. Luu committed
Dec 17, 2025 at 02:30 UTC
4bf6c2366818a1eeae257e06ec337039e6895f13
5 files changed
+10
-21
.buildkite/test_areas/e2e_integration.yaml
+1
-18
@@ -32,6 +32,7 @@ steps:
32
- label: Prime-RL Integration (2 GPUs)
33
timeout_in_minutes: 30
34
optional: true
35
+ soft_fail: true
36
num_gpus: 2
37
working_dir: "/vllm-workspace"
38
source_file_dependencies:
@@ -39,21 +40,3 @@ steps:
40
- .buildkite/scripts/run-prime-rl-test.sh
41
commands:
42
- bash .buildkite/scripts/run-prime-rl-test.sh
42
-
43
-- label: DeepSeek V2-Lite Async EPLB Accuracy
44
- timeout_in_minutes: 60
45
- gpu: h100
46
- optional: true
47
- num_gpus: 4
48
- working_dir: "/vllm-workspace"
49
- commands:
50
- - bash .buildkite/scripts/scheduled_integration_test/deepseek_v2_lite_ep_async_eplb.sh 0.25 1319 8030
51
-
52
-- label: Qwen3-Next-80B-A3B-Instruct MTP Async EPLB Accuracy
53
- timeout_in_minutes: 60
54
- gpu: h100
55
- optional: true
56
- num_gpus: 4
57
- working_dir: "/vllm-workspace"
58
- commands:
59
- - bash .buildkite/scripts/scheduled_integration_test/qwen3_next_mtp_async_eplb.sh 0.8 1319 8040
.buildkite/test_areas/lm_eval.yaml
+2
-2
@@ -9,7 +9,7 @@ steps:
9
- vllm/model_executor/layers/quantization
10
autorun_on_main: true
11
commands:
12
- - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-small.txt --tp-size=1
12
+ - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-small.txt
13
14
- label: LM Eval Large Models (4 GPUs)(A100)
15
gpu: a100
@@ -43,4 +43,4 @@ steps:
43
- csrc/
44
- vllm/model_executor/layers/quantization
45
commands:
46
- - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-blackwell.txt --tp-size=1
46
+ - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-blackwell.txt
.buildkite/test_areas/lora.yaml
+2
@@ -22,6 +22,8 @@ steps:
22
# FIXIT: find out which code initialize cuda before running the test
23
# before the fix, we need to use spawn to test it
24
- export VLLM_WORKER_MULTIPROC_METHOD=spawn
25
+ # Alot of these tests are on the edge of OOMing
26
+ - export PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True
27
# There is some Tensor Parallelism related processing logic in LoRA that
28
# requires multi-GPU testing for validation.
29
- pytest -v -s -x lora/test_chatglm3_tp.py
.buildkite/test_areas/models_basic.yaml
+2
@@ -9,6 +9,7 @@ steps:
9
source_file_dependencies:
10
- vllm/
11
- tests/models/test_initialization.py
12
+ - tests/models/registry.py
13
commands:
14
# Run a subset of model initialization tests
15
- pytest -v -s models/test_initialization.py::test_can_initialize_small_subset
@@ -20,6 +21,7 @@ steps:
21
source_file_dependencies:
22
- vllm/model_executor/models/
23
- tests/models/test_initialization.py
24
+ - tests/models/registry.py
25
commands:
26
# Only when vLLM model source is modified - test initialization of a large
27
# subset of supported models (the complement of the small subset in the above
.buildkite/test_areas/pytorch.yaml
+3
-1
@@ -13,7 +13,9 @@ steps:
13
# tests covered elsewhere.
14
# Use `find` to launch multiple instances of pytest so that
15
# they do not suffer from https://github.com/vllm-project/vllm/issues/28965
16
- - "find compile/ -maxdepth 1 -name 'test_*.py' -exec pytest -s -v {} \\;"
16
+ # However, find does not normally propagate error codes, so we combine it with xargs
17
+ # (using -0 for proper path handling)
18
+ - "find compile/ -maxdepth 1 -name 'test_*.py' -print0 | xargs -0 -n1 -I{} pytest -s -v '{}'"
19
20
- label: PyTorch Fullgraph Smoke Test
21
timeout_in_minutes: 30