From 82c51298dbc30227d341239821d84f4d7a85aa7b Mon Sep 17 00:00:00 2001 From: Syed Jafri Date: Mon, 31 Aug 2026 21:58:53 +0000 Subject: [PATCH 1/2] fix: update instance type to g6e for deepseek integ tests --- .../inference/cli/test_cli_jumpstart_inference.py | 2 +- .../inference/cli/test_cli_jumpstart_v12_fields.py | 2 +- .../inference/sdk/test_sdk_jumpstart_inference.py | 2 +- .../inference/sdk/test_sdk_jumpstart_v12_fields.py | 2 +- test/integration_tests/init/test_jumpstart_creation.py | 4 ++-- 5 files changed, 6 insertions(+), 6 deletions(-) diff --git a/test/integration_tests/inference/cli/test_cli_jumpstart_inference.py b/test/integration_tests/inference/cli/test_cli_jumpstart_inference.py index 6d7e2c3c..8bec3118 100644 --- a/test/integration_tests/inference/cli/test_cli_jumpstart_inference.py +++ b/test/integration_tests/inference/cli/test_cli_jumpstart_inference.py @@ -34,7 +34,7 @@ def test_js_create(runner, js_endpoint_name): "--namespace", NAMESPACE, "--version", VERSION, "--model-id", "deepseek-llm-r1-distill-qwen-1-5b", - "--instance-type", "ml.g5.8xlarge", + "--instance-type", "ml.g6e.2xlarge", "--endpoint-name", js_endpoint_name, ]) assert result.exit_code == 0, result.output diff --git a/test/integration_tests/inference/cli/test_cli_jumpstart_v12_fields.py b/test/integration_tests/inference/cli/test_cli_jumpstart_v12_fields.py index 2ec0572a..2e33354f 100644 --- a/test/integration_tests/inference/cli/test_cli_jumpstart_v12_fields.py +++ b/test/integration_tests/inference/cli/test_cli_jumpstart_v12_fields.py @@ -34,7 +34,7 @@ def test_create_with_v12_fields(runner, endpoint_name): "--namespace", NAMESPACE, "--version", VERSION, "--model-id", "deepseek-llm-r1-distill-qwen-1-5b", - "--instance-type", "ml.g5.8xlarge", + "--instance-type", "ml.g6e.2xlarge", "--endpoint-name", endpoint_name, "--accept-eula", "True", # v1.2 new fields diff --git a/test/integration_tests/inference/sdk/test_sdk_jumpstart_inference.py b/test/integration_tests/inference/sdk/test_sdk_jumpstart_inference.py index 98bf9269..2593cfec 100644 --- a/test/integration_tests/inference/sdk/test_sdk_jumpstart_inference.py +++ b/test/integration_tests/inference/sdk/test_sdk_jumpstart_inference.py @@ -14,7 +14,7 @@ REGION = "us-east-2" ENDPOINT_NAME = "js-sdk-integration-" + get_time_str() -INSTANCE_TYPE = "ml.g5.8xlarge" +INSTANCE_TYPE = "ml.g6e.2xlarge" MODEL_ID = "deepseek-llm-r1-distill-qwen-1-5b" TIMEOUT_MINUTES = 20 diff --git a/test/integration_tests/inference/sdk/test_sdk_jumpstart_v12_fields.py b/test/integration_tests/inference/sdk/test_sdk_jumpstart_v12_fields.py index 5f77f9d4..58f34e16 100644 --- a/test/integration_tests/inference/sdk/test_sdk_jumpstart_v12_fields.py +++ b/test/integration_tests/inference/sdk/test_sdk_jumpstart_v12_fields.py @@ -23,7 +23,7 @@ def endpoint_obj(): return HPJumpStartEndpoint( metadata=metadata, model=Model(model_id="deepseek-llm-r1-distill-qwen-1-5b", accept_eula=True), - server=Server(instance_type="ml.g5.8xlarge"), + server=Server(instance_type="ml.g6e.2xlarge"), sage_maker_endpoint=SageMakerEndpoint(name=ENDPOINT_NAME), replicas=1, max_deploy_time_in_seconds=5400, diff --git a/test/integration_tests/init/test_jumpstart_creation.py b/test/integration_tests/init/test_jumpstart_creation.py index 04bb1f2c..2546635b 100644 --- a/test/integration_tests/init/test_jumpstart_creation.py +++ b/test/integration_tests/init/test_jumpstart_creation.py @@ -89,7 +89,7 @@ def test_configure_jumpstart(runner, js_endpoint_name, test_directory): result = runner.invoke( configure, [ "--model-id", "deepseek-llm-r1-distill-qwen-1-5b", - "--instance-type", "ml.g5.8xlarge", + "--instance-type", "ml.g6e.2xlarge", "--endpoint-name", js_endpoint_name ], catch_exceptions=False ) @@ -98,7 +98,7 @@ def test_configure_jumpstart(runner, js_endpoint_name, test_directory): # Verify configuration was saved correctly expected_config = { "model_id": "deepseek-llm-r1-distill-qwen-1-5b", - "instance_type": "ml.g5.8xlarge", + "instance_type": "ml.g6e.2xlarge", "endpoint_name": js_endpoint_name } assert_config_values("./", expected_config) From 82b6bd2cc1b148cf87c9735b831c34169d5aa680 Mon Sep 17 00:00:00 2001 From: Syed Jafri Date: Mon, 31 Aug 2026 23:57:01 +0000 Subject: [PATCH 2/2] fix: use smaller model so tests run on g5.8xlarge --- .../inference/cli/test_cli_jumpstart_inference.py | 4 ++-- .../inference/cli/test_cli_jumpstart_v12_fields.py | 4 ++-- .../inference/sdk/test_sdk_jumpstart_inference.py | 4 ++-- .../inference/sdk/test_sdk_jumpstart_v12_fields.py | 4 ++-- test/integration_tests/init/test_jumpstart_creation.py | 8 ++++---- 5 files changed, 12 insertions(+), 12 deletions(-) diff --git a/test/integration_tests/inference/cli/test_cli_jumpstart_inference.py b/test/integration_tests/inference/cli/test_cli_jumpstart_inference.py index 8bec3118..5097eb0d 100644 --- a/test/integration_tests/inference/cli/test_cli_jumpstart_inference.py +++ b/test/integration_tests/inference/cli/test_cli_jumpstart_inference.py @@ -33,8 +33,8 @@ def test_js_create(runner, js_endpoint_name): result = runner.invoke(js_create, [ "--namespace", NAMESPACE, "--version", VERSION, - "--model-id", "deepseek-llm-r1-distill-qwen-1-5b", - "--instance-type", "ml.g6e.2xlarge", + "--model-id", "huggingface-llm-phi-2", + "--instance-type", "ml.g5.8xlarge", "--endpoint-name", js_endpoint_name, ]) assert result.exit_code == 0, result.output diff --git a/test/integration_tests/inference/cli/test_cli_jumpstart_v12_fields.py b/test/integration_tests/inference/cli/test_cli_jumpstart_v12_fields.py index 2e33354f..cc9af8e1 100644 --- a/test/integration_tests/inference/cli/test_cli_jumpstart_v12_fields.py +++ b/test/integration_tests/inference/cli/test_cli_jumpstart_v12_fields.py @@ -33,8 +33,8 @@ def test_create_with_v12_fields(runner, endpoint_name): result = runner.invoke(js_create, [ "--namespace", NAMESPACE, "--version", VERSION, - "--model-id", "deepseek-llm-r1-distill-qwen-1-5b", - "--instance-type", "ml.g6e.2xlarge", + "--model-id", "huggingface-llm-phi-2", + "--instance-type", "ml.g5.8xlarge", "--endpoint-name", endpoint_name, "--accept-eula", "True", # v1.2 new fields diff --git a/test/integration_tests/inference/sdk/test_sdk_jumpstart_inference.py b/test/integration_tests/inference/sdk/test_sdk_jumpstart_inference.py index 2593cfec..6926ebb5 100644 --- a/test/integration_tests/inference/sdk/test_sdk_jumpstart_inference.py +++ b/test/integration_tests/inference/sdk/test_sdk_jumpstart_inference.py @@ -14,8 +14,8 @@ REGION = "us-east-2" ENDPOINT_NAME = "js-sdk-integration-" + get_time_str() -INSTANCE_TYPE = "ml.g6e.2xlarge" -MODEL_ID = "deepseek-llm-r1-distill-qwen-1-5b" +INSTANCE_TYPE = "ml.g5.8xlarge" +MODEL_ID = "huggingface-llm-phi-2" TIMEOUT_MINUTES = 20 POLL_INTERVAL_SECONDS = 30 diff --git a/test/integration_tests/inference/sdk/test_sdk_jumpstart_v12_fields.py b/test/integration_tests/inference/sdk/test_sdk_jumpstart_v12_fields.py index 58f34e16..1e028608 100644 --- a/test/integration_tests/inference/sdk/test_sdk_jumpstart_v12_fields.py +++ b/test/integration_tests/inference/sdk/test_sdk_jumpstart_v12_fields.py @@ -22,8 +22,8 @@ def endpoint_obj(): return HPJumpStartEndpoint( metadata=metadata, - model=Model(model_id="deepseek-llm-r1-distill-qwen-1-5b", accept_eula=True), - server=Server(instance_type="ml.g6e.2xlarge"), + model=Model(model_id="huggingface-llm-phi-2", accept_eula=True), + server=Server(instance_type="ml.g5.8xlarge"), sage_maker_endpoint=SageMakerEndpoint(name=ENDPOINT_NAME), replicas=1, max_deploy_time_in_seconds=5400, diff --git a/test/integration_tests/init/test_jumpstart_creation.py b/test/integration_tests/init/test_jumpstart_creation.py index 2546635b..cb4c02e7 100644 --- a/test/integration_tests/init/test_jumpstart_creation.py +++ b/test/integration_tests/init/test_jumpstart_creation.py @@ -88,8 +88,8 @@ def test_configure_jumpstart(runner, js_endpoint_name, test_directory): configure = init.configure result = runner.invoke( configure, [ - "--model-id", "deepseek-llm-r1-distill-qwen-1-5b", - "--instance-type", "ml.g6e.2xlarge", + "--model-id", "huggingface-llm-phi-2", + "--instance-type", "ml.g5.8xlarge", "--endpoint-name", js_endpoint_name ], catch_exceptions=False ) @@ -97,8 +97,8 @@ def test_configure_jumpstart(runner, js_endpoint_name, test_directory): # Verify configuration was saved correctly expected_config = { - "model_id": "deepseek-llm-r1-distill-qwen-1-5b", - "instance_type": "ml.g6e.2xlarge", + "model_id": "huggingface-llm-phi-2", + "instance_type": "ml.g5.8xlarge", "endpoint_name": js_endpoint_name } assert_config_values("./", expected_config)