# Beyond the KV cache: Training and serving i14's Laplace 3B linear attention model on AWS Trainium2 with native PyTorch

> **Publication**: aitalky (Independent AI Journalism)
> **Canonical Citation URL**: https://aitalky.vercel.app/news/beyond-the-kv-cache-training-and-serving-i14s-laplace-3b-linear-attention-model-on
> **Original Source**: [AWS ML Blog](https://aws.amazon.com/blogs/machine-learning/beyond-the-kv-cache-training-and-serving-i14s-laplace-3b-linear-attention-model-on-aws-trainium2-with-native-pytorch/)
> **Author**: Kevin Shabahang (undefined)
> **Published Date**: Tuesday, October 20, 2026
> **Category**: INDUSTRY
> **Tags**: AI, AWS ML Blog, industry

## Executive Summary
Learn how i14 trained and served Laplace 3B, a linear attention model, on AWS Trainium2 using native PyTorch. Its recurrent state stays fixed at 29.5 MB per sequence regardless of context length, avoiding the growing key-value cache of transformer models and...



## Full Coverage
Most large language models (LLMs) use transformer architectures with softmax attention. During inference, their key-value (KV) cache grows with the conversation or document context. This can reduce concurrency, increase serving cost, and force agentic applications to discard useful history

i14 designed Laplace 3B as a linear attention alternative. Its recurrent state stays at 29.5 MB per sequence, independent of context length. Working with AWS, i14 used native PyTorch support through TorchNeuron Native to bring this custom architecture to AWS Trainium2 . This preserved its Gated Linear Attention (GLA) design and Muon optimizer

On one trn2.48xlarge instance, Laplace 3B trained at about 120,000 tokens per second. The model served 11,544 output tokens per second across 1,280 concurrent sequences. This post explains the model design, the changes i14 made for Trainium2, and the training and inference results

---
*Generated for AI agents and LLM citation by aitalky (https://aitalky.vercel.app).*