biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[CI] Speed up model unit tests in CI (#24253)

Browse Source 
Signed-off-by: Andrew Feldman <afeldman@redhat.com>
This commit is contained in:
afeldman-nm
2025-09-12 13:36:50 -04:00
committed by GitHub
parent 9d2a44606d
commit c8c42597ab
6 changed files with 123 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
tests/models/language/generation/test_common.py
View File

@@ -38,7 +38,7 @@ AITER_MODEL_LIST = [
[
pytest.param(
"bigscience/bloom-560m", # bloom - testing alibi slopes
marks=[pytest.mark.core_model],
marks=[pytest.mark.core_model, pytest.mark.slow_test],
),
pytest.param(
"openai-community/gpt2", # gpt2
@@ -49,7 +49,10 @@ AITER_MODEL_LIST = [
pytest.param("EleutherAI/pythia-70m"), # gpt_neox
pytest.param(
"google/gemma-1.1-2b-it", # gemma
marks=[pytest.mark.core_model, pytest.mark.cpu_model],
marks=[
pytest.mark.core_model, pytest.mark.cpu_model,
pytest.mark.slow_test
],
),
pytest.param(
"zai-org/chatglm3-6b", # chatglm (text-only)
@@ -70,14 +73,17 @@ AITER_MODEL_LIST = [
),
pytest.param(
"microsoft/phi-2", # phi
marks=[pytest.mark.core_model],
marks=[pytest.mark.core_model, pytest.mark.slow_test],
),
pytest.param(
"Qwen/Qwen-7B-Chat", # qwen (text-only)
),
pytest.param(
"Qwen/Qwen2.5-0.5B-Instruct", # qwen2
marks=[pytest.mark.core_model, pytest.mark.cpu_model],
marks=[
pytest.mark.core_model, pytest.mark.cpu_model,
pytest.mark.slow_test
],
),
pytest.param(
"Qwen/Qwen3-8B", # qwen (text-only)

5
tests/models/language/pooling/test_classification.py
View File

@@ -11,7 +11,10 @@ from vllm.platforms import current_platform
"model",
[
pytest.param("jason9693/Qwen2.5-1.5B-apeach",
marks=[pytest.mark.core_model, pytest.mark.cpu_model]),
marks=[
pytest.mark.core_model, pytest.mark.cpu_model,
pytest.mark.slow_test
]),
],
)
@pytest.mark.parametrize("dtype",

7
tests/models/language/pooling/test_embedding.py
View File

@@ -19,7 +19,7 @@ from ...utils import check_embeddings_close
# model code with bidirectional attention.
# [Decoder-only]
pytest.param("BAAI/bge-multilingual-gemma2",
marks=[pytest.mark.core_model]),
marks=[pytest.mark.core_model, pytest.mark.slow_test]),
pytest.param(
"intfloat/e5-mistral-7b-instruct",
# CPU v1 doesn't support sliding window
@@ -29,7 +29,10 @@ from ...utils import check_embeddings_close
# [Encoder-only]
pytest.param(
"BAAI/bge-base-en-v1.5",
marks=[pytest.mark.core_model, pytest.mark.cpu_model],
marks=[
pytest.mark.core_model, pytest.mark.cpu_model,
pytest.mark.slow_test
],
),
pytest.param("sentence-transformers/all-MiniLM-L12-v2"),
pytest.param("intfloat/multilingual-e5-small"),