DLC Framework Support Policy


Created On: September 30, 2022
Last Updated: March 12, 2025


Supported Framework Versions

For more details see the Framework Support Policy.

Framework Current version CUDA version GitHub GA End of patch
PyTorch 2.6 - 12.6 for Training DLCs
- 12.4 for Inference DLCs
2025-01-29 2026-01-29
PyTorch 2.5 12.4 2024-10-29 2025-10-29
PyTorch 2.4 12.4 2024-07-24 2025-07-24
PyTorch 2.3 12.1 2024-04-24 2025-04-24
TensorFlow 2.18 - 12.5 for Training DLCs
- 12.2 for Inference DLCs
2025-01-24 2026-01-24

 

Unsupported Framework Versions

Versions listed in this table will appear for 2 years past their support date.

Framework Current version CUDA version GitHub GA End of patch
PyTorch 2.2 - 12.1 for Training DLCs
- 11.8 for Inference DLCs
2024-01-30 2025-01-30
PyTorch 2.1 - 12.1 for Training DLCs
- 11.8 for Inference DLCs
2023-10-04 2024-10-04
PyTorch 2.0 - 12.1 for Training DLCs
- 11.8 for Training and Inference DLCs
2023-03-15 2024-03-15
PyTorch 1.13 11.7 2022-10-28 2024-10-28
PyTorch 1.12
– 11.3 for SageMaker DLC
- 11.6 for EC2/ECS/EKS DLC
2022-07-01 2023-07-01
PyTorch 1.11
- 11.3 for SageMaker DLC
- 11.5 for EC2/ECS/EKS DLC
2022-03-10 2023-03-10
TensorFlow 2.16 - 12.3 for Training DLCs
- 12.2 for Inference DLCs
2024-03-07 2025-03-07
TensorFlow 2.14 11.8 2023-09-26 2024-09-26
TensorFlow 2.13 11.8 2023-07-19 2024-07-19

TensorFlow
2.12 11.8 2023-03-23 2024-03-23

TensorFlow
2.11 11.2 2022-11-18 2023-11-18

TensorFlow
2.10 11.2 2022-09-06 2023-09-06
TensorFlow 2.9 11.2 2022-05-17 2023-05-17
TensorFlow 2.8 11.2 2022-02-02 2023-02-02
TensorFlow 2.7 11.2 2021-11-04 2022-11-04

 

Updates

  • 2024-08-14 Updated the version format from major.minor.patch to major.minor to reduce confusion.
  • 2023-10-30 Extend support for PT 1.13 by one year, as it is the last 1.x version released by PyTorch.
  • 2023-10-01 End of support for MXNet and Elastic Inference

 

Unsupported Frameworks

From 1st Oct 2023, MXNet, Elastic Inference frameworks are not supported because of lack of upstream support on those frameworks. Please use one of our supported frameworks as per section “Supported Framework Versions” in this document.