Redundant tf-nightly-gpu and tensorflow-gpu being removed
71 views
Skip to first unread message
Austin Anderson
Jan 24, 2023, 11:16:49 PM1/24/23
to TensorFlow Developers
Hi everyone,
TL;DR If you are using tf-nightly-gpu or tensorflow-gpu, switch to tf-nightly or tensorflow. The packages are identical for TensorFlow versions newer than 2.2. We are going to stop publishing these redundant *-gpu packages immediately. TF 2.12 will not have a -gpu package.
The tf-nightly-gpu and tensorflow-gpu packages will no longer be published. For a long time, ever since TensorFlow 2.1 (see release notes) these packages have been identical to their non-gpu variants: tf-nightly-gpu is the same package as tf-nightly, and tensorflow-gpu is the same package as tensorflow. All of these packages support GPU functionality with Nvidia CUDA. As part of a long-delayed tech debt cleanup, we're going to stop publishing these redundant packages.
If you are using "*-gpu" packages in your CI or day-to-day workflow, please switch over to the regular versions. These still have GPU support and there is no benefit to using the "*-gpu" packages unless you are using a TensorFlow 1.X release, in which case you do not need to change anything.
Here's what to expect:
Austin Anderson
TensorFlow DevInfra
TL;DR If you are using tf-nightly-gpu or tensorflow-gpu, switch to tf-nightly or tensorflow. The packages are identical for TensorFlow versions newer than 2.2. We are going to stop publishing these redundant *-gpu packages immediately. TF 2.12 will not have a -gpu package.
The tf-nightly-gpu and tensorflow-gpu packages will no longer be published. For a long time, ever since TensorFlow 2.1 (see release notes) these packages have been identical to their non-gpu variants: tf-nightly-gpu is the same package as tf-nightly, and tensorflow-gpu is the same package as tensorflow. All of these packages support GPU functionality with Nvidia CUDA. As part of a long-delayed tech debt cleanup, we're going to stop publishing these redundant packages.
If you are using "*-gpu" packages in your CI or day-to-day workflow, please switch over to the regular versions. These still have GPU support and there is no benefit to using the "*-gpu" packages unless you are using a TensorFlow 1.X release, in which case you do not need to change anything.
Here's what to expect:
- Starting today, tf-nightly-gpu will stop updating.
- Starting today, tensorflow-gpu will stop updating. We will not release any future TensorFlow patch releases for tensorflow-gpu.
- We're publishing a final version of tf-nightly-gpu and tensorflow-gpu that will warn you to install tf-nightly or tensorflow instead. This package will be published within the next few days and will interrupt any installation that does not specify an old version, so "pip install tensorflow-gpu" will raise an error.
- We are not deleting any currently-published packages. You can still run "pip install tensorflow-gpu==[an old version]"
Austin Anderson
TensorFlow DevInfra
Reply all
Reply to author
Forward
0 new messages