Integ test updates

aws · May 17, 2024 · ed9d005 · ed9d005
1 parent 4bb5522
commit ed9d005
Show file tree

Hide file tree

Showing 2 changed files with 2 additions and 2 deletions.
diff --git a/src/sagemaker/serve/model_server/tgi/server.py b/src/sagemaker/serve/model_server/tgi/server.py
@@ -74,7 +74,7 @@ def _invoke_tgi_serving(self, request: object, content_type: str, accept: str):
         """Placeholder docstring"""
         try:
             response = requests.post(
-                f"http://{get_docker_host()}:8080/generate",
+                f"http://{get_docker_host()}:8080/invocations",
                 data=request,
                 headers={"Content-Type": content_type, "Accept": accept},
                 timeout=600,

diff --git a/tests/integ/sagemaker/serve/test_serve_tei.py b/tests/integ/sagemaker/serve/test_serve_tei.py
@@ -107,7 +107,7 @@ def test_tei_sagemaker_endpoint(sagemaker_session, model_builder, model_input):
     with timeout(minutes=SERVE_SAGEMAKER_ENDPOINT_TIMEOUT):
         try:
             logger.info("Deploying and predicting in SAGEMAKER_ENDPOINT mode...")
-            predictor = model.deploy(instance_type="ml.g5.12xlarge", initial_instance_count=2)
+            predictor = model.deploy(instance_type="ml.g5.2xlarge", initial_instance_count=1)
             logger.info("Endpoint successfully deployed.")
             predictor.predict(model_input)
             assert predictor is not None