Universal Sentence Encoder Failing to Deploy on SageMaker Notebook instance]

[singularity014]

Hello I am trying to use TensorFlow Serving for using Universal Sentence Encoder using
Sagemaker's NoteBook instance.
It throws the error -
" The primary container for production variant AllTraffic did not pass the ping health check. Please check CloudWatch logs for this endpoint..! "
""

To reproduce
A clear, step-by-step set of instructions to reproduce the bug.

from sagemaker import get_execution_role
import sagemaker
role = get_execution_role()
from sagemaker.tensorflow.serving import Model, Predictor

sagemaker_model = Model(
model_data = 's3://naister-platform-models/MUSE_TAR/universal-sentence-encoder.tar.gz',
role = role
)

predictor = sagemaker_model.deploy(
initial_instance_count=1,
instance_type='ml.t2.medium',
endpoint_name="MuseEndpointLg"
)

input = ['I am a good guy.', 'I am a bad guy, very bad.']
output = predictor.predict(input)

Expected behavior
Similarity score between the two input sentences.

Screenshots or logs

" The primary container for production variant AllTraffic did not pass the ping health check. Please check CloudWatch logs for this endpoint..! "
""

[metrizable]

@singularity014 Thanks for filing the issue. I have a few questions:

• Have you checked the CloudWatch logs for the endpoint? And, if so, can you share?
• I see you are using an ml.t2.medium. Is your tar.gz of reasonable size?
• From here, your endpoint has 4 minutes to move to healthy state "by consistently responding with 200s". Is it possible that start-up is taking longer than this? The logs may provide some insight. Trying a different instance type may yield results, too.

[singularity014]

@metrizable thanks for addressing the question.

• I am getting following error log 'No saved model found'

.

• The tar.gz model is of 318 MB in size.

• Yes, the enpoint creation process was taking very long :
  predictor = sagemaker_model.deploy(
  initial_instance_count=1,
  instance_type='ml.t2.medium',
  endpoint_name="MuseEndpointLg"
  )

It took almost 15 minutes to run this , before it threw the error.

[laurenyu]

sorry for the delayed response here. what are the contents of s3://naister-platform-models/MUSE_TAR/universal-sentence-encoder.tar.gz?