Two architectures that didn't help small-model agent memory on a free T4

📰 Dev.to · Atharva Vichare

A one-week experiment on LongMemEval_S with Qwen3-4B, 50 stratified questions, and honest arithmetic.

Published 20 Apr 2026