AutoRound

AI summary

AutoRound is a quantization toolkit for Large Language Models and Vision Language Models that utilizes sign gradient descent to achieve high accuracy at 2–4 bit widths.

AI named AutoRound from November 2025 to September 2026.

AutoRound in AI’s own words

Question: I want to quantize my model to run faster and cheaper. What's the best model quantization library or toolkit?

Google AI ModeSep 6, 2026

AutoRound (by Intel) uses signed gradient descent to optimize low-bit weights

Ask Google AI Mode yourself (opens in a new tab)

Markets

ML Deployment & Inference Optimization Tools
Edge AI Model Optimization Tools