Skip to content

Commit ea900bf

Browse files
committed
Move to G5 for integ test
1 parent 4c336dd commit ea900bf

File tree

1 file changed

+5
-9
lines changed

1 file changed

+5
-9
lines changed

tests/integ/sagemaker/serve/test_serve_tei.py

Lines changed: 5 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -15,15 +15,15 @@
1515
import pytest
1616
from sagemaker.serve.builder.schema_builder import SchemaBuilder
1717
from sagemaker.serve.builder.model_builder import ModelBuilder, Mode
18-
import tests.integ
18+
1919
from tests.integ.sagemaker.serve.constants import (
2020
HF_DIR,
2121
PYTHON_VERSION_IS_NOT_310,
2222
SERVE_SAGEMAKER_ENDPOINT_TIMEOUT,
2323
)
2424

2525
from tests.integ.timeout import timeout
26-
from tests.integ.utils import cleanup_model_resources, gpu_list, retry_with_instance_list
26+
from tests.integ.utils import cleanup_model_resources
2727
import logging
2828

2929
logger = logging.getLogger(__name__)
@@ -89,14 +89,10 @@ def model_builder(request):
8989

9090

9191
@pytest.mark.skipif(
92-
PYTHON_VERSION_IS_NOT_310,
93-
tests.integ.test_region() in tests.integ.TRAINING_NO_P2_REGIONS
94-
and tests.integ.test_region() in tests.integ.TRAINING_NO_P3_REGIONS,
95-
reason="no ml.p2 or ml.p3 instances in this region",
92+
PYTHON_VERSION_IS_NOT_310
9693
)
97-
@retry_with_instance_list(gpu_list(tests.integ.test_region()))
9894
@pytest.mark.parametrize("model_builder", ["model_builder_model_schema_builder"], indirect=True)
99-
def test_tei_sagemaker_endpoint(sagemaker_session, model_builder, model_input, **kwargs):
95+
def test_tei_sagemaker_endpoint(sagemaker_session, model_builder, model_input):
10096
logger.info("Running in SAGEMAKER_ENDPOINT mode...")
10197
caught_ex = None
10298

@@ -111,7 +107,7 @@ def test_tei_sagemaker_endpoint(sagemaker_session, model_builder, model_input, *
111107
try:
112108
logger.info("Deploying and predicting in SAGEMAKER_ENDPOINT mode...")
113109
predictor = model.deploy(
114-
instance_type=kwargs["instance_type"], initial_instance_count=2
110+
instance_type="ml.g5.12xlarge", initial_instance_count=2
115111
)
116112
logger.info("Endpoint successfully deployed.")
117113
predictor.predict(model_input)

0 commit comments

Comments
 (0)