All Posts

Announcing the deprecation of the NeuronX Deep Learning Inference API path_trace function

Neuron release 2.26.0 is the last release supporting parallel_model_trace. This NxD Inference function will be deprecated in the next version of the Neuron SDK in favor of the ModelBuilder.trace() method, which provides a more robust and flexible approach for tracing and compiling models for Neuron devices, enabling more advanced features such as weight layout optimization support, as well as other quality-of-life and stability improvements for SPMD tracing.

For customers directly invoking parallel_model_trace, they can now use ModelBuilderV2 APIs. For more details on these APIS, see ModelBuilderV2 API Reference. For customers that are directly using models in NxDI, there is no impact since NxDI models are already built on MBv1 which has no issues.

Read more ...

Announcing End of Support for PyTorch NeuronX version 2.6 starting next release

Neuron release 2.26 will be the last release that will include support for PyTorch NeuronX version 2.6. Future Neuron releases will not include support for PyTorch NeuronX version 2.6. Current users of PyTorch NeuronX version 2.6 are advised to upgrade to PyTorch NeuronX 2.7 or 2.8. See Neuron release artifacts for more details on supported versions.

Read more ...

Upcoming changes to Neuron driver 2.21 support for Inf1 starting Neuron 2.26 release

Starting with Neuron Release 2.26, Neuron driver versions above 2.21 will only support non-Inf1 instances (such as Trn1, Inf2, or other instance types). For Inf1 instance users, Neuron driver versions < 2.21 will remain supported with regular security patches.

Inf1 instance users are advised to pin the Neuron driver version to 2.21.* in their installation script. Refer to the Neuron Driver release [2.22.2.0] for detailed instructions on pinning the Neuron Driver.

Read more ...

Neuron no longer supports nki_jit API in PyTorch Neuron starting this release

Starting with Neuron Release 2.24, torch_neuronx.nki_jit API in torch-neuronx package is no longer supported.

I currently use nki_jit in my PyTorch models. What do I do?

Read more ...

Announcing end of support for Transformers NeuronX library starting in Neuron 2.26 release

Starting from Neuron Release 2.24, Transformers NeuronX library is in maintenance mode. transformers-neuronx releases will now only address critical security issues. In Neuron Release 2.26, Neuron will end support for transformers-neuronx.

Current users of transformers-neuronx are advised to migrate to NeuronX Distributed Inference.

Read more ...

Announcing end of support for NKI block dimension starting next release

Neuron release 2.24 will be the last release to include support for the NKI block dimension in NKI tensor creation routines. Starting with this release, using the block dimension will generate EOS warnings. In the next release (Neuron Release 2.25), these warnings will be upgraded to errors.

Customers are recommended to refer to the NKI Block Dimension Migration Guide for detailed instructions on updating their code.

Read more ...

Announcing end of support for Llama 3.2 Meta checkpoint

Starting with Neuron Release 2.24, the mllama 3.2 Meta checkpoint API is no longer be supported.

I currently use the mllama 3.2 Meta checkpoint in my applications. What do I do?

Read more ...

Announcing end of support for Beta PyTorch NeuronCore Placement APIs starting next release

Neuron Release 2.24 is the last release to support the Beta PyTorch NeuronCore Placement APIs.

Customers using Beta PyTorch NeuronCore Placement APIs are recommended to migrate to using generally available (GA) PyTorch Neuron Core Placement APIs. Please refer to the PyTorch Neuron documentation for guidance on using the supported functionality. Any models using the beta APIs will need to be updated to use the generally available APIs.

Read more ...

Announcing end of support XLA_USE_BF16 and XLA_DOWNCAST_BF16 environment variables starting next release

Neuron Release 2.24 will be the last release to support the following environment variables:

XLA_USE_BF16

Read more ...

Announcing End of Support for Tensorflow Neuron Inf1 SSD300 tutorial starting next release

Neuron release 2.24 will be the last release that will include support for Tensorflow Neuron Inf1 SSD300 tutorial. Future Neuron releases will not include support for Tensorflow Neuron Inf1 SSD300 tutorial due to security issues.

Read more ...

Announcing End of Support for PyTorch NeuronX version 2.5 starting next release

Neuron release 2.24 will be the last release that will include support for PyTorch NeuronX version 2.5. Future Neuron releases will not include support for PyTorch NeuronX version 2.5.

Current users of PyTorch NeuronX version 2.5 are advised to upgrade to PyTorch NeuronX 2.6 or 2.7. Please see release artifacts for more details on supported versions.

Read more ...

Neuron no longer supports jax_neuronx.nki_call API in jax-neuronx starting this release

  • 15 May 2025
  • en

Neuron Release 2.23 no longer supports jax_neuronx.nki_call API in jax-neuronx package.

For a full list of features that require jax-neuronx, please see JAX NeuronX Known Issues.

Read more ...

Neuron no longer supports base Ubuntu 20 operating system starting this release

Neuron Release 2.23 no longer includes support for base Ubuntu 20.04 operating system.

Customers using Ubuntu 20.04 are required to migrate their workloads to Ubuntu 22.04 or another supported operating system. Please refer to Neuron DLAMI User Guide for guidance on Neuron supported operating systems.

Read more ...

Neuron no longer supports Tensorflow 2.8 and 2.9 starting this release

Starting with Neuron Release 2.23, Neuron no longer supports for TensorFlow-Neuron 2.8 and 2.9 versions.

Current users of those versions are advised to migrate to latest TensorFlow version (2.10). For a list of supported versions, please see Release Content.

Read more ...

Neuron no longer supports NeMo Megatron starting this release

Starting with Neuron release 2.23, Neuron no longer supports NeMo Megatron.

All users of AWS Neuron Reference for NeMo Megatron are requested to migrate their training workloads to NxD Training. Please refer to Neuron NeMo Megatron to NeuronX Distributed Training Migration Guide for guidance.

Read more ...

Announcing migration of NxD Core inference examples from NxD Core repository to NxD Inference repository starting this release

Starting with Neuron Release 2.23, the following models and modules in NxD Core inference examples are now only available through NxD Inference package:

Llama

Read more ...

Announcing maintenance mode for Transformers NeuronX library starting next release

Starting from Neuron release 2.24, Transformers NeuronX library is entering maintenance mode. Future releases of transformers-neuronx will address critical security issues only and we will gradually end support.

Current users of transformers-neuronx are advised to migrate to NeuronX Distributed Inference.

Read more ...

Announcing end of support for torch_neuronx.nki_jit API in torch-neuronx starting next release

Neuron Release 2.23 will be the last release to include support for torch_neuronx.nki_jit API in torch-neuronx package.

Customers using torch_neuronx.nki_jit API are recommended to switch invocations to directly call functions annotated with nki.jit API.

Read more ...

Announcing end of support for mllama 3.2 Meta Checkpoint API starting next release

Neuron Release 2.23 will be the last release to include support for the mllama 3.2 Meta checkpoint API. In the next release (Neuron 2.24), Neuron will end support.

All previously converted checkpoints will continue to function without disruption. Customers’ existing workflows and converted models remain fully operational. For new checkpoint conversions, the HuggingFace solution provides equivalent functionality. Customers are recommended to use HuggingFace’s official conversion script, available here: Hugging Face Conversion Script

Read more ...

Announcing end of support for XLA_USE_BF16 and XLA_DOWNCAST_BF16 starting next release

Starting with Neuron Release 2.23, Neuron will begin phasing out support for the XLA_DOWNCAST_BF16 and XLA_USE_BF16 environment variables. In this release, usage of these variables will trigger warnings. Neuron will end support in a subsequent release, aligned with the torch-xla maintenance schedule.

Customers are recommended to migrate to automatic mixed-precision or use model.to(torch.bfloat16) command to convert their model to BF16 format. For detailed migration guidance, please refer to Migration From XLA_USE_BF16/XLA_DOWNCAST_BF16.

Read more ...

Neuron no longer supports PyTorch 2.1 starting this release

Starting with Neuron Release 2.22, Neuron no longer includes support for PyTorch 2.1, its associated Deep Learning Containers (DLCs), and Deep Learning AMIs (DLAMIS).

We recommend that all customers using PyTorch 2.1, related DLCs, and DLAMIS upgrade to the latest supported PyTorch version. For more information on supported versions, please refer to Release Content.

Read more ...

Neuron no longer supports PyTorch 1.13 starting this release

Starting with Neuron Release 2.22, Neuron no longer supports PyTorch 1.13, its associated Deep Learning Containers (DLCs), and Deep Learning AMIs (DLAMIS) for Trn1, Trn2, and Inf2 instances.

We recommend that all customers using PyTorch 1.13, related DLCs, and DLAMIS on Trn2, Trn1, and Inf2 instances upgrade to the latest supported PyTorch version. For more information on supported versions, please refer to Release Content.

Read more ...

Neuron no longer includes support for Ubuntu20 DLCs and DLAMIs starting this release

Starting with Neuron Release 2.22, Neuron no longer includes offerings for Ubuntu20 Deep Learning Containers (DLCs) and Deep Learning AMIs (DLAMIs).

Customers using Ubuntu20 DLCs and DLAMIs should migrate to newer versions based on Ubuntu22 or Amazon Linux 2023. For customers who need to continue using Ubuntu20, you can create custom AMIs based on the Ubuntu20 base image and install Neuron components manually. Please see Neuron Containers FAQ and Neuron DLAMI User Guide.

Read more ...

Neuron no longer includes support for Neuron DET tool starting with this release

Starting with Neuron Release 2.22, Neuron no longer supports the Neuron Distributed Event Tracing (NDET/neuron-det) tool (Neuron Distributed Event Tracing (NDET) User Guide).

We recommend customers transition to the Neuron Profiler 2.0 for debugging runtime hangs and issues in large-scale settings. This tool offers the same runtime function level traces with improved ease of use and optimized performance. For more information about the Neuron Profiler 2.0, see Neuron Profiler 2.0 (Beta) User Guide.

Read more ...

Neuron no longer includes Python 3.8 support starting this release

Starting with Neuron Release 2.22, Neuron no longer includes support for Python 3.8 as it has its reached end-of-life status.

To avoid security issues and bugs, current users of Python 3.8 are advised to migrate to a Neuron supported Python version (3.9, 3.10, or 3.11) as Neuron no longer supports Python 3.8. For a list of supported Python versions according to Neuron package, please see Release Content.

Read more ...

Announcing end of support for jax_neuronx.nki_call API in jax-neuronx from starting next release

Starting with Neuron Release 2.23, Neuron will end support for jax_neuronx.nki_call API in jax-neuronx package.

For a full list of features that require jax-neuronx, please see JAX NeuronX Known Issues.

Read more ...

Announcing end of support for Tensorflow 2.8 and 2.9 starting next release

Starting with Neuron Release 2.23, Neuron will end support for TensorFlow 2.8 and 2.9. Future Neuron releases will not include support for Tensorflow-Neuron 2.8 and 2.9 versions.

Current users of those versions are advised to migrate to latest TensorFlow version (2.10). For a list of supported versions, please see Release Content.

Read more ...

Announcing end of support for Neuron support for NeMo Megatron starting next release

Starting with Neuron Release 2.23, Neuron will end support for NeMo Megatron.

We recommend all users of NeMo Megatron to migrate their training workloads to NxD Training. Please refer to Neuron NeMo Megatron to NeuronX Distributed Training Migration Guide for guidance.

Read more ...

PyTorch Neuron versions 1.9 and 1.10 no longer supported

Starting with Neuron Release 2.21, Neuron no longer supports torch-neuron 1.9 and 1.10 versions. Current users of torch-neuron 1.9 and 1.10 versions are advised to migrate to the latest torch-neuron supported version. Please see Release Content.

Read more ...

Announcing migration of NxD Core examples from NxD Core repository to NxD Inference repository in next release

Neuron Release 2.21 will be the last release to include NxD Core repository inference examples under the NxD Core repository: aws-neuron/neuronx-distributed. Starting with Neuron Release 2.21, the models and modules in NxD Core inference examples are now available through NxD Inference package. We recommend customers to update their applications to use examples from the NxD Inference repository. See NxD Inference Overview

In Neuron Release 2.22, the NxD Core inference samples will only reside under the NxD Inference repository. Current users are advised to start using samples/tutorials under the NxD Inference repository: aws-neuron/neuronx-distributed-inference.

Read more ...

Announcing end of support for Ubuntu20 DLCs and DLAMIs

Starting with Neuron Release 2.21, AWS Neuron will begin phasing out support for Ubuntu20 Deep Learning Containers (DLCs) and Deep Learning AMIs (DLAMIs). Neuron 2.21 will be the last release to provide bug fixes, and by Neuron 2.22, these offerings will no longer be available.

We recommend that all customers using Ubuntu20 DLCs and DLAMIs migrate to newer versions based on Ubuntu22 or Amazon Linux 2023. For customers who need to continue using Ubuntu20, you can create custom AMIs based on the Ubuntu20 base image and install Neuron components manually. Please see Neuron Containers FAQ and Neuron DLAMI User Guide.

Read more ...

Announcing end of support for Python 3.8 in future releases

Due to Python 3.8 reaching its end-of-life status, future Neuron releases will no longer include support for this version.

To avoid security issues and bugs, current users of Python 3.8 are advised to migrate to a Neuron supported Python version (3.9, 3.10, or 3.11) as Neuron will no longer support Python 3.8. For a list of supported Python versions according to Neuron package, please see Release Content.

Read more ...

Announcing end of support for PyTorch 2.1 starting next release

Neuron Release 2.21 is the last release to support PyTorch 2.1, its associated Deep Learning Containers (DLCs), and Deep Learning AMIs (DLAMIS).

We recommend that all customers using PyTorch 2.1, related DLCs, and DLAMIS upgrade to the latest supported PyTorch version. For more information on supported versions, please refer to Release Content.

Read more ...

Announcing end of support for PyTorch 1.13 starting next release

Neuron Release 2.21 is the last release to support PyTorch 1.13, its associated Deep Learning Containers (DLCs), and Deep Learning AMIs (DLAMIS) for Trn1, Trn2, and Inf2 instances.

We recommend that all customers using torch-neuron 1.13, related DLCs, and DLAMIS on Trn2, Trn1, and Inf2 instances upgrade to the latest supported PyTorch version. For more information on supported versions, please refer to Release Content.

Read more ...

Announcing end of support for Neuron DET tool starting next release

Neuron Release 2.21 will be the last release to support the Neuron Distributed Event Tracing (NDET/neuron-det) tool (Neuron Distributed Event Tracing (NDET) User Guide).

We recommend all customers using the NDET tool for debugging runtime hangs/issues in large-scale settings transition to the Neuron Profiler 2.0. This tool offers the same runtime function level traces with improved ease of use and optimized performance. For more information on Neuron Profiler 2.0, please refer to the Neuron Profiler 2.0 (Beta) User Guide.

Read more ...

‘neurondevice’ resource name in Neuron Device K8s plugin no longer supported

Starting with Neuron release 2.20, Neuron no longer supports resource name ‘neurondevice’.

Neuron device plugin is a Neuron Software component that gets installed in Kubernetes environment. The resource name ‘neurondevice’ enables customers to allocate devices to the Neuron K8s container.

Read more ...

‘neuron-device-version’ field in neuron-monitor no longer supported

Starting with Neuron release 2.20, Neuron no longer supports the field ‘neuron-device-version’ in neuron-monitor.

Customers who are using the field ‘neuron-device-version’ will instead need to use ‘instance_type’ field in the ‘instance_info’ section and the ‘neuroncore_version’ field to obtain neuron device information.

Read more ...

Tensorflow-Neuron 1.x no longer supported

Starting with Neuron release 2.20, Neuron no longer supports Tensorflow-Neuron 1.x. Current users of those versions are advised to migrate to latest tensorflow-neuron version 2.10.1. Please see TensorFlow Neuron for more details.

Read more ...

Neuron Runtime no longer supports Amazon Linux 2 (AL2)

Starting from Neuron release 2.20, the Neuron Runtime (aws-neuronx-runtime-lib) will no longer support Amazon Linux 2 (AL2). The Neuron Driver (aws-neuronx-dkms) is now the only Neuron package that supports Amazon Linux 2.

This change introduces the following constraint:

Read more ...

Announcing maintenance mode for torch-neuron 1.9 and 1.10 versions

Starting with Neuron release 2.20, torch-neuron 1.9 and 1.10 versions will enter maintenance mode. Future Neuron releases will not include support for torch-neuron 1.9 and 1.10 versions. Current users of torch-neuron 1.9 and 1.10 versions are advised to migrate to torch-neuron 1.13.

Read more ...

Announcing Name Change for Neuron Component

Starting with Neuron release 2.20, the name of the following Neuron component will change as follows:

Package name

Read more ...

Neuron support for MxNet enters maintenance mode

Starting with Neuron release 2.19, Neuron support for MxNet (mxnet-neuron) is entering maintenance mode.

Future releases of mxnet-neuron will address critical security issues only and we will gradually end support. Current users of mxnet-neuron are advised to migrate to PyTorch NeuronX or TensorFlow NeuronX.

Read more ...

Announcing end of support for ‘neurondevice’ resource name in Neuron Device K8s plugin

Neuron release 2.19 will be the last release to include resource name ‘neurondevice’.

Neuron device plugin is a Neuron Software component that gets installed in Kubernetes environment. The resource name ‘neurondevice’ enables customers to allocate devices to the Neuron K8s container.

Read more ...

Announcing end of support for ‘neuron-device-version’ field in neuron-monitor

Neuron release 2.19 will be the last release to include the field ‘neuron-device-version’ in neuron-monitor.

In future releases, customers who are using the field ‘neuron-device-version’ will instead need to use ‘instance_type’ field in the ‘instance_info’ section and the ‘neuroncore_version’ field to obtain neuron device information.

Read more ...

Announcing end of support for Tensorflow-Neuron 1.x

Neuron release 2.19 will be the last release to support Tensorflow-Neuron 1.x. Future Neuron releases will not include support for Tensorflow-Neuron 1.x versions. Current users of those versions are advised to migrate to latest tensorflow-neuron version 2.10.1.

Read more ...

Announcing end of support for Probuf versions <= 3.19 for PyTorch NeuronX, NeuronX Distributed, and Transformers NeuronX libraries

Neuron release 2.19 will be the last release that will include Probuf <= 3.19 support for PyTorch NeuronX, NeuronX Distributed, and Transformers NeuronX libraries. Future Neuron releases will not include Probuf <= 3.19 support for PyTorch NeuronX.

Current PyTorch NeuronX, NeuronX Distributed, or Transformers NeuronX users using Probuf <= 3.19 are advised to migrate to latest supported Probuf version.

Read more ...

Announcing end of support for Neuron Runtime support of Amazon Linux 2 (AL2)

Neuron release 2.19 will be the last release that will include Neuron Runtime support for Amazon Linux 2 . Future Neuron releases will not include Neuron Runtime support for Amazon Linux 2.

Current users using Amazon Linux 2 are advised to migrate to Amazon Linux 2023 (AL2023) or Ubuntu 20/22.

Read more ...

Announcing end of support for Neuron virtual environments in AWS Deep Learning AMI (Amazon Linux 2)

Neuron release 2.18.2 will be the last release that will include support for the following virtual environments in AWS Deep Learning AMI (Amazon Linux 2):

aws_neuron_pytorch_p38: PyTorch 1.13, Python 3.8

Read more ...

Announcing end of support for Neuron Release 2.18.0 Deep Learning AMIs

We are announcing end of support for Neuron release 2.18.0 Deep Learning AMIs. DLAMIs released between March 26,2024 (2024-03-26) and April 10, 2024 (2024-04-10) were shipped without the audit package. The following are the affected DLAMIs:

Deep Learning AMI Neuron (Ubuntu 22.04) 20240401

Read more ...

Tensorflow-Neuron 1.x enters maintenance mode

Starting with Neuron release 2.18, Tensorflow-Neuron 1.x is entering maintenance mode. Future releases of Tensorflow-Neuron 1.x will address critical security issues only and we will gradually end support. Current users of those versions are advised to migrate to latest tensorflow-neuron version 2.10.1.

Read more ...

PyTorch NeuronX version 2.0 (Beta) no longer supported

Neuron release 2.17 no longer supports PyTorch NeuronX version 2.0 (Beta).

Current users of PyTorch NeuronX version 2.0 are advised to migrate to PyTorch NeuronX 2.1 (Beta).

Read more ...

Announcing Name Change for Neuron Components

Starting with Neuron release 2.16, the name of the following Neuron components will change as follows:

Package name

Read more ...

Announcing end-of-support for NeuronX Distributed Training Samples in Neuron Samples Repository

Neuron release 2.16 will be the last release to include support for NeuronX Distributed Training Samples (Llama-2, GPT-NeoX 20B, and GPT-NeoX 6.9B) under the AWS Neuron Samples Github repository.

In future releases, NeuronX Distributed samples will reside under the NeuronX Distributed Github repository. Current users are advised to start using samples under the NeuronX Distributed repository for all NeuronX Distributed tutorials.

Read more ...

Announcing end-of-support for Neuron Containers with Runtime 1.x

Neuron release 2.3 was the last release to support Neuron Runtime 1.x (neuron-rtd). Current users of Neuron DLC/DLAMI with Neuron Runtime 1.x are required to update their image to support latest Neuron Runtime versions. For instructions, see the Setup Guide.

Read more ...

Announcing End of Support for PyTorch NeuronX version 2.0 (beta)

Neuron release 2.16 will be the last release that will include beta support for PyTorch NeuronX version 2.0 (beta). Future Neuron releases will not include support for PyTorch NeuronX version 2.0.

Current users of PyTorch NeuronX version 2.0 are advised to upgrade to PyTorch NeuronX 2.1 (beta).

Read more ...

Announcing End of Support for PyTorch Neuron version 1.10

Neuron release 2.16 will be the last release that will include support for PyTorch Neuron version 1.10. Future Neuron releases will not include support for PyTorch Neuron version 1.10.

Current users of PyTorch Neuron version 1.10 are advised to migrate to latest supported PyTorch Neuron version.

Read more ...

Announcing End of Support for nemo option-argument

Neuron release 2.15 will be the last release that will include support for nemo option-argument in the existing –distribution_strategy compiler option. Future releases will not include Neuron support for nemo option-argument. Users are advised to migrate to the new llm-training option-argument.

Read more ...

Announcing End Of Support for OPT example in Transformers NeuronX

Neuron release 2.15 will be the last release that will include OPT example in Transformers NeuronX.

Read more ...

Announcing end-of-support for --model-type=transformer-inference compiler flag

Starting with Neuron release 2.14, the --model-type=transformer-inference compiler flag is deprecated.

Neuron SDK users using the --model-type=transformer-inference compiler flag are highly encouraged to migrate to the --model-type=transformer compiler flag.

Read more ...

Announcing end of support for torch-neuron version 1.9

Neuron release 2.13 will be the last release that will include support for torch-neuron version 1.9. Future Neuron releases will not include support for torch-neuron version 1.9.

Current users of torch-neuron version 1.9 are advised to migrate to latest supported torch-neuron version.

Read more ...

AWS Neuron reference for Megatron-LM no longer supported

Neuron release 2.13 no longer includes support for AWS Neuron reference for Megatron-LM.

Current Neuron Megatron-LM users are required to migrate to AWS Neuron reference for NeMo Megatron or Neuron Distributed.

Read more ...

Announcing end of support for AWS Neuron reference for Megatron-LM

Neuron release 2.12 will be the last release that will include support for AWS Neuron reference for Megatron-LM. Future releases will not include Neuron support for Megatron-LM.

Current Neuron Megatron-LM users are advised to migrate to AWS Neuron reference for NeMo Megatron or Neuron Distributed.

Read more ...

AWS Deep Learning AMIs now available with Neuron 2.12 version

We are happy to announce that the following Deep Learning AMIs are now available with latest Neuron Version 2.12.

You can see more about the AMIs at the following URLs

Read more ...

Announcing end of support for Python 3.7

Neuron release 2.12 will be the last release that will include support for Python 3.7 . Future Neuron releases will not include support for Python 3.7

Current users using Python 3.7 are advised to migrate to latest supported Python version. (Python 3.10 )

Read more ...

Announcing end of support for Ubuntu 18

Neuron release 2.12 will be the last release that will include support for Ubuntu 18 . Future Neuron releases will not include support for Ubuntu 18

Current users using Ubuntu 18 are advised to migrate to Ubuntu 20 version.

Read more ...

AWS Deep Learning AMIs now available with Neuron 2.10 version

We are happy to announce that the following Deep Learning AMIs are now available with latest Neuron Version 2.10. These DLAMIs now support all the Neuron EC2 instances including Inf1, Inf2, Trn1/Trn1n.

You can access the AMIs at the following URLs

Read more ...

Announcing end of support for tensorflow-neuron versions 2.7

Neuron release 2.10 will be the last release that will include tensorflow-neuron versions 2.7. Future Neuron releases will not include tensorflow-neuron versions 2.7

Current users of those versions are advised to migrate to latest tensorflow-neuron version.

Read more ...

Announcing end of support for mxnet-neuron versions 1.5

Neuron release 2.10 will be the last release that will include mxnet-neuron versions 1.5. Future Neuron releases will not include mxnet-neuron versions 1.5

Current users of those versions are advised to migrate to latest mxnet-neuron version.

Read more ...

PyTorch 1.13 Deep Learning Container for Inf2 & Trn1/Trn1n now available for SageMaker

We are happy to announce that an updated Deep Learning Container that supports PyTorch 1.13 and Neuron 2.9.1 versions is now available for Sagemaker Training.

For more information see Neuron Containers

Read more ...

Introducing Neuron packaging and installation changes for Inf1 customers

Starting with Neuron release 2.5, Neuron introduces changes in Neuron packages and installation instructions for Inf1, the following Neuron packages will change names:

New name

Read more ...

Announcing end of support for tensorflow-neuron versions 2.5 and 2.6

Neuron release 2.5 will be the last release that will include tensorflow-neuron versions 2.5 and 2.6. Future Neuron releases will not include tensorflow-neuron versions 2.5 and 2.6.

Current users of those versions are advised to migrate to latest tensorflow-neuron version.

Read more ...

Announcing end of support for torch-neuron versions 1.7 and 1.8

Neuron release 2.5 will be the last release that will include torch-neuron versions 1.7 and 1.8. Future Neuron releases will not include torch-neuron versions 1.7 and 1.8.

Current users of those versions are advised to migrate to latest torch-neuron version.

Read more ...

Neuron GPG key for Ubuntu installation has expired

GPG, or GNU Privacy Guard, is a public key cryptography implementation. This allows for the secure transmission of information between parties and can be used to verify that the origin of a message is genuine.

The GPG key for the Neuron repository (https://apt.repos.neuron.amazonaws.com/GPG-PUB-KEY-AMAZON-AWS-NEURON.PUB) is installed on the Ubuntu (Canonical) server, the key was uploaded originally with an expiry date of three (3) years, which has expired on 11/10/22.

Read more ...

Amazon SageMaker now supports Trn1 training jobs

We are happy to announce that Amazon SageMaker now supports running training jobs on ml.trn1 instance types.

For more information see Distributed Training with PyTorch Neuron on Trn1 instances

Read more ...

Introducing AWS Deep Learning AMI Neuron PyTorch

We are happy to announce that Deep Learning AMI (DLAMI) with pre-installed PyTorch Neuron (torch-neuronx) is now available, for more information see:

AWS Deep Learning AMI Neuron PyTorch 1.11 (Amazon Linux 2)

Read more ...

Introducing the first release of Neuron 2.x enabling EC2 Trn1 General Availability (GA)

Neuron release 2.3 is the first release of Neuron 2.x that enables GA of the new EC2 Trn1 instances. Neuron release 2.3 extends the latest release of Neuron 1.x (Neuron 1.19.2), adding support for Deep Learning training on the AWS Trainium chips.

Starting with Neuron release 2.3, developers can run Deep Learning training workloads on Trn1 instances, saving training costs by up to 50% over equivalent GPU-based EC2 instances, while achieving the highest training performance in the AWS cloud for popular NLP models. Neuron 2.x introduces new capabilities and major architectural updates to support training neural-networks with the Trn1 instances.

Read more ...

Introducing Packaging and installation changes

Starting with Neuron release 2.3, Neuron introduces changes in Neuron packages and installation instructions.

New Neuron packages

Read more ...

Introducing New Neuron GitHub Repositories

Starting with Neuron release 2.3, Neuron Github repositories will be migrated to the new AWS Neuron GitHub Organization.

The new AWS Neuron GitHub Organization will include the Neuron SDK GitHub repository and will include the following additional new GitHub repositories:

Read more ...

Announcing Neuron Runtime 1.x (neuron-rtd) end-of-support

Starting with Neuron release 2.3, Neuron components like Neuron System Tools and Neuron Driver will no longer support Neuron Runtime 1.x.

In addition, starting with Neuron release 2.3, the AWS Neuron Runtime Proto GitHub and AWS Neuron Driver GitHub repositories will no longer be supported.

Read more ...

End of support for torch-neuron version 1.5

Starting with Neuron 1.19.0 release, torch-neuron 1.5 will no longer be supported, and no further releases of torch-neuron version 1.5 will be issued. Current users of torch-neuron version 1.5 are advised to migrate to latest torch-neuron version.

Read more ...

End of support for tensorflow-neuron versions 2.1, 2.2, 2.3 and 2.4

Starting with Neuron 1.18.0 release, tensorflow-neuron versions 2.1, 2.2, 2.3 and 2.4 will no longer be supported, and no further releases of tensorflow-neuron versions 2.1, 2.2, 2.3 and 2.4 will be issued. Current users of those versions are advised to migrate to latest tensorflow-neuron version.

Read more ...

Announcing end of support for torch-neuron version 1.5 starting with Neuron 1.19.0 release

Starting with Neuron 1.19.0 release, torch-neuron version 1.5 will no longer be supported. Last release of torch-neuron version 1.5 will be issued as part of Neuron 1.18.0 release. Current users of those versions are advised to migrate to latest torch-neuron version.

Read more ...

Announcing end of support for NEURONCORE_GROUP_SIZES starting with Neuron 1.20.0 release

Starting with Neuron SDK 1.20.0, NEURONCORE_GROUP_SIZES environment variable will no longer be supported. Setting NEURONCORE_GROUP_SIZES environment variable will no longer affect applications behavior. Current customers using NEURONCORE_GROUP_SIZES environment variable are advised to use NEURON_RT_VISIBLE_CORES environment variable or NEURON_RT_NUM_CORES environment variable instead.

See 10/27/2021 - End of support for NeuronCore Groups (NCG), NeuronX Runtime Configuration and Migrate your application to Neuron Runtime 2.x (libnrt.so) for more information.

Read more ...

Previous Announcements

02/17/2022 - tensorflow-neuron versions 2.1, 2.2, 2.3 and 2.4 enter maintenance mode

10/27/2021 - Introducing Neuron Runtime 2.x (libnrt.so)

Read more ...