Skip to content

Commit ed9d005

Browse files
committed
Integ test updates
1 parent 4bb5522 commit ed9d005

File tree

2 files changed

+2
-2
lines changed

2 files changed

+2
-2
lines changed

src/sagemaker/serve/model_server/tgi/server.py

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -74,7 +74,7 @@ def _invoke_tgi_serving(self, request: object, content_type: str, accept: str):
7474
"""Placeholder docstring"""
7575
try:
7676
response = requests.post(
77-
f"http://{get_docker_host()}:8080/generate",
77+
f"http://{get_docker_host()}:8080/invocations",
7878
data=request,
7979
headers={"Content-Type": content_type, "Accept": accept},
8080
timeout=600,

tests/integ/sagemaker/serve/test_serve_tei.py

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -107,7 +107,7 @@ def test_tei_sagemaker_endpoint(sagemaker_session, model_builder, model_input):
107107
with timeout(minutes=SERVE_SAGEMAKER_ENDPOINT_TIMEOUT):
108108
try:
109109
logger.info("Deploying and predicting in SAGEMAKER_ENDPOINT mode...")
110-
predictor = model.deploy(instance_type="ml.g5.12xlarge", initial_instance_count=2)
110+
predictor = model.deploy(instance_type="ml.g5.2xlarge", initial_instance_count=1)
111111
logger.info("Endpoint successfully deployed.")
112112
predictor.predict(model_input)
113113
assert predictor is not None

0 commit comments

Comments
 (0)