Skip to main content

A light-weight library of utilities for TensorFlow 2/Keras.

Project description

tfu

A small library of many useful utilities for Tensorflow training workflows.

Filtering Visible Devices

In some instances, you may want to filter what devices are visible to Tensorflow. For example, you may have multiple GPUS, but would prefer to use only a single GPU instance when training your model. tf_utils includes some convenience methods to assist with this.

Selecting a single GPU:

visible_devices = tfu.devices.select_gpu(0)

Selecting a subset of GPUs:

visible_devices = tfu.devices.select_gpu([0, 1])

Training Strategies

When a training strategy is needed, in particular OneDeviceStrategy or MirroredDeviceStrategy, tf_utils provides some convenience functions to create these strategies automatically and filter out unused devices. Some examples are provided below.

Use the CPU only via a OneDeviceStrategy:

strategy = tfu.strategy.cpu(0)

Use a single GPU via the OneDeviceStrategy:

strategy = tfu.strategy.gpu(0)

Use multiple GPUs via the MirroredDeviceStrategy:

strategy = tfu.strategy.gpu([0, 1])

Dynamic Memory Growth

Tensorflow has the ability to use dynamic memory allocation, rather than allocating all of the memory on the GPU at once. Enabling dynamic memory allocation allows you to not only monitor memory usage of your models during training, but it also grants you the ability to run multiple models on a single GPU instance. All device selection and strategy functions within tf_utils support this feature via the use_dynamic_memory flag.

Enabling dynamic memory growth via device selection:

visible_devices = tfu.devices.select_gpu(0, use_dynamic_memory=True)

Enabling dynamic memory growth via strategy creation:

strategy = tfu.strategy.gpu(0, use_dynamic_memory=True)

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

tf-utilities-0.1.9.tar.gz (10.7 kB view details)

Uploaded Source

Built Distribution

tf_utilities-0.1.9-py3-none-any.whl (11.4 kB view details)

Uploaded Python 3

File details

Details for the file tf-utilities-0.1.9.tar.gz.

File metadata

  • Download URL: tf-utilities-0.1.9.tar.gz
  • Upload date:
  • Size: 10.7 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/4.0.2 CPython/3.10.6

File hashes

Hashes for tf-utilities-0.1.9.tar.gz
Algorithm Hash digest
SHA256 6b0efeb1c8fdb79b4c229063724c40ad6e9a8a3ce2e26526aed48816b497d827
MD5 50ebeb465945acaae74cd66cef71456b
BLAKE2b-256 d18a0017aea4edc6872e97e0a0f8269c2b2d1fc1ae030fe56820b5f2d5d6df39

See more details on using hashes here.

File details

Details for the file tf_utilities-0.1.9-py3-none-any.whl.

File metadata

File hashes

Hashes for tf_utilities-0.1.9-py3-none-any.whl
Algorithm Hash digest
SHA256 8aa84ea67a9236098dd71cdd7325b86cde5d102da4cc1edeb722e092159f8a3e
MD5 fb59bea99a4369a5b4085fb95651b308
BLAKE2b-256 bbc0e23d048fc80a8c3b3046b660ae61bbb256fa061cd3fa5057c71f39c5d8ae

See more details on using hashes here.

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page