Added tutorial for using torchserve on aws sagemaker #2671
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/tutorials/2671
Note: Links to docs will display an error until the docs builds have been completed. This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
@pytorchbot label "docathon-h2-2023" |
| | TorchServe is easy to use. It comes with a convenient CLI to deploy locally and is easy to package into a container and scale out with Amazon SageMaker or Amazon EKS. With default handlers for common problems such as image classification, object detection, image segmentation, and text classification, you can deploy with just a few lines of code—no more writing lengthy service handlers for initialization, preprocessing, and post-processing. TorchServe is open-source, which means it's fully open and extensible to fit your deployment needs. | ||
|
|
||
| To get started on how to use TorchServe you can refer to this tutorial: `TorchServe QuickStart <https://pytorch.org/serve/getting_started.html>`_ | ||
|
|
There was a problem hiding this comment.
Please add the user manual links of using TorchServe on SM:
There was a problem hiding this comment.
SageMaker has 2 different endpoints. Their deployment is slightly different. Please include this information.
- single model
- multi-model
There was a problem hiding this comment.
Please add the user manual links of using TorchServe on SM:
* https://docs.aws.amazon.com/sagemaker/latest/dg/deploy-models-frameworks-torchserve.html * https://docs.aws.amazon.com/sagemaker/latest/dg/large-model-inference-tutorials-torchserve.html
Hello lxning,
Thanks for the review, I have already added these links in the reference at the end of the tutorial.
| #. Create a compressed tar.gz file out of the densenet161.mar file, because Amazon SageMaker expects models to be in a tar.gz file. | ||
|
|
||
| .. code:: shell | ||
|
|
||
| tar cvfz $model_file_name.tar.gz densenet161.mar |
There was a problem hiding this comment.
- You can skip this extra step by using torchserve-model-archiver --archive-format tgz.
- For large model, we recommend using torchserve-model-archiver --archive-format no-archive by leveraging SM uncompressed model artifact feature (current ly only available on SageMaker single model endpoint) (see details: com/sagemaker/latest/dg/large-model-inference-tutorials-torchserve.html)
|
|
||
| aws s3 cp $model_file_name.tar.gz s3://{bucket_name}/{prefix}/model | ||
|
|
||
| Creating an Amazon ECR registry |
There was a problem hiding this comment.
This step is needed ONLY if you are going to BYOD or BYOC, otherwise it is not needed.
| Metrics | ||
| ~~~~~~~~ | ||
|
|
||
| TorchServe supports both system level and model level metrics. You can enable metrics in either log format mode or Prometheus mode through the environment variable TS_METRICS_MODE. You can use the TorchServe central metrics config file metrics.yaml to specify the types of metrics to be tracked, such as request counts, latency, memory usage, GPU utilization, and more. By referring to this file, you can gain insights into the performance and health of the deployed models and effectively monitor the TorchServe server's behavior in real-time. For more detailed information, see the `TorchServe metrics documentation <https://github.com/pytorch/serve/blob/master/docs/metrics.md#torchserve-metrics>`_. You can access TorchServe metrics logs that are similar to the StatsD format through the Amazon CloudWatch log filter. The following is an example of a TorchServe metrics log: |
There was a problem hiding this comment.
SageMaker does not support prometheus format. User can only use regex search TorchServe metrics log.
prometheus format not yet supported to view torchserve metric logs boto/boto3#3437
…ure for uncompressed models
|
Hello @svekars @sekyondaMeta, just a gentle reminder to review my PR kindly. Should I remove the |
Fixes #2345
Added tutorial on how to use
torchserveonAWS Sagemaker.The tutorial focuses on features of AWS Sagemaker and other AWS services which we can use to serve PyTorch model rather than emphasizing various features provided by torchserve. Although I have provided external links to tutorials wherever we can do more things using more torchserve features like handler customization in torch-model-archiver.
Checklist
cc @msaroufim @agunapal @sekyondaMeta @svekars @carljparker @NicolasHug @kit1980 @subramen