# Amazon — Profiling TorchNeuron (Part 1): System-level profiling on AWS Trainium.

- Company: Amazon (amazon.com)
- Announced: 2026-10-21T16:04:43+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://huggingface.co/blog/amazon/profiling-torchneuron-part-1
- Record: https://forck.live/items/26022-profiling-torchneuron-part-1-system-level-profiling-on-aws-trainium

Amazon's TorchNeuron team published the first part of a guide teaching ML practitioners how to profile inference performance on AWS Trainium, starting from no prior profiling experience. The walkthrough shows how to configure torch.profiler with the Neuron backend, read the key_averages() table to see which operations consume Neuron device time, and follow a single operation through Neuron Explorer from framework dispatch to hardware execution. Neuron integrates with PyTorch through the PrivateUse1 extension point, so the torch_neuronx package registers it as a native backend without a dedicated dispatch key. Profiling uses NeuronConfig to select data streams and NeuronCores and NeuronProfiler to export a framework trace alongside Neuron system traces. The guide profiles three implementations of RMSNorm — individual PyTorch operations, the built-in torch.rms_norm(), and the naive version under torch.compile — using identical inputs and schedules so results are comparable. Reproduction scripts are in the torch-neuronx examples directory.

## Evidence

Verbatim from https://huggingface.co/blog/amazon/profiling-torchneuron-part-1:

> In Part 1 of this series, you will learn how to systematically analyze inference performance and identify where the time is spent.

---

Record: https://forck.live/items/26022-profiling-torchneuron-part-1-system-level-profiling-on-aws-trainium
Catalogue: https://forck.live/llms.txt
Current issue: https://forck.live/feed.md