How can I connect to to inference machine in order to debug the custom image?

[andrew-aladjev]

Hello, I've derived my custom docker image from your sagemaker image. Dockerfile is the following:

FROM 763104351884.dkr.ecr.us-east-1.amazonaws.com/pytorch-inference:2.5.1-cpu-py311-ubuntu22.04-sagemaker

RUN conda install -y ffmpeg

I built this image, uploaded it to ecr and used new custom image ARN in image_uri of PyTorchModel as follows:

model = PyTorchModel(
  model_data = f's3://{SAGEMAKER_BUCKET_NAME}/{MODEL_ARCHIVE_NAME}',
  entry_point = 'inference.py',
  role = role,
  sagemaker_session = session,
  image_uri = '...',
  framework_version = '2.5',
  py_version = 'py311'
)

model.deploy(
  instance_type = '...',
  initial_instance_count = 1,
  endpoint_name = '...'
)

It works: sagemaker deployed my model, but inference code is still not working: torchaudio can't see ffmpeg backend. Now I want to debug my machine. In order to debug it I want to make any kind of terminal connection. But I am unable to do it, because I can't find any ability to do it.

I've clicked everything in Amazon UI: Amazon SageMaker AI, Inference menu, Notebook menu, Jupiter menu. I have even opened fluffy toy "Studio" and I can't see anything. I see just endpoint without any ability to connect to the related machine(s). What should I do? For now I am thinking to move away from Amazon.