https://trentenwen.com/ Notes from the trenches of putting intelligence on devices that don't have much. en https://trentenwen.com/posts/running-llms-on-4-chip.html https://trentenwen.com/posts/running-llms-on-4-chip.html How I got an 8 KB keyword-spotting model below 4 KB by moving quantization noise into training. Mon, 15 Dec 2025 00:00:00 GMT