AWS blog shows how to train Qwen3 on Trainium with torchtitan
What happened
On 22 October 2026, the AWS Machine Learning Blog published a guide to training Qwen3 on AWS Trainium with torchtitan. According to the post, on a single trn2.48xlarge instance (16 Trainium2 chips, 64 logical NeuronCores), torchtitan can register Trainium as a native PyTorch device through TorchNeuron. The report gives no further details such as training scale, duration or performance comparisons.
Written by AI from the reports · updated 1 h ago
Timeline
Follow the reports to see the event from each side.
- AWS Machine Learning BlogTrain Qwen3 on AWS Trainium with torchtitan
Train Qwen3 on AWS Trainium with torchtitan three ways on a single trn2.48xlarge: the Hugging Face Transformers v5 modeling backend, torchtitan's native Qwen3, and NKI mixture-of-experts kernels. The same torchrun launcher and config structure work unchanged across all three paths, including loading and saving published Hugging Face safetensors.
Heat of this event
Not enough continuous observations yet to draw a trend.