Skip to content

Changelog

Changelog for the Amazon Linux 2023-based TensorFlow Serving SageMaker inference images (2.20-gpu-py312-cu129-amzn2023-sagemaker and 2.20-cpu-py312-amzn2023-sagemaker).

TensorFlow Serving 2.20 — 2026-08-20

Tags: 2.20-gpu-py312-cu129-amzn2023-sagemaker · 2.20-cpu-py312-amzn2023-sagemaker · 2.20.0-gpu-py312-cu129-amzn2023-sagemaker · 2.20.0-cpu-py312-amzn2023-sagemaker

Bundled versions: TensorFlow Serving 2.20.0 · Python 3.12 · CUDA 12.9.1 · cuDNN 9.24.0.43 · nginx 1.30.3 · njs 0.9.9

Highlights

  • Initial release of the TensorFlow inference DLC on Amazon Linux 2023 and Python 3.12
  • TensorFlow Serving 2.20.0 for SageMaker AI inference (CPU and GPU variants)
  • CUDA 12.9.1 with cuDNN 9.24.0.43 on the GPU variant, with automatic CUDA forward compatibility on older host drivers
  • nginx 1.30.3 with the njs 0.9.9 module fronting the model server on port 8080
  • Multi-model endpoints supported (SAGEMAKER_MULTI_MODEL=true), including per-model inference.py handlers
  • Server-side batching, thread-pool sizing, and multiple model server processes configurable via SAGEMAKER_TFS_* environment variables
  • The tensorflow framework wheel, EFA, OpenMPI, and NCCL are intentionally excluded — only the TensorFlow Serving gRPC stubs ship (tensorflow-serving-api-gpu on GPU, tensorflow-serving-api on CPU), keeping the image small