To elaborate, user could tune environment parameters such as
SAGEMAKER_GUNICORN_WORKERS, SAGEMAKER_TFS_INSTANCE_COUNT,
SAGEMAKER_TFS_INTER_OP_PARALLELISM,
SAGEMAKER_TFS_INTRA_OP_PARALLELISM and etc.
In this way container will be started with multiple of gunicorn
workers and they are responsible for serving requests to different
tensorflow model servers parallelly. Thus, throughput will be
improved based on different parameters tuned.
By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.
Implement new container environment which could expose additional parameters to help user optimize for throughput and fine tune container performance.
Co-authored-by: Liang Ma liangmaa@amazon.com
Issue #, if available:
Description of changes:
To elaborate, user could tune environment parameters such as SAGEMAKER_GUNICORN_WORKERS, SAGEMAKER_TFS_INSTANCE_COUNT, SAGEMAKER_TFS_INTER_OP_PARALLELISM, SAGEMAKER_TFS_INTRA_OP_PARALLELISM and etc. In this way container will be started with multiple of gunicorn workers and they are responsible for serving requests to different tensorflow model servers parallelly. Thus, throughput will be improved based on different parameters tuned.
By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.