Radar ·
mini-AGI explores continual learning with 8 GB of VRAM
Published on September 22, 2026, the account describes a model trained from scratch on a notebook with 8 GB of VRAM, using batch size 1, disk-paged weights, and learning-rate adjustments to reduce forgetting.
Published on September 22, 2026, the account about the mini-AGI repository describes a model trained from scratch on a notebook with 8 GB of VRAM. Its data flow uses batch size 1, pages weights from disk, and adjusts learning rates in an effort to reduce forgetting during continual learning.
The implementation is presented as material for examining online learning and model training with limited memory. To verify the scope of the claims, consult the original repository and post, inspect the code and reported settings, and compare the documented results; the available summary gives no performance metrics.
Original source: mini-AGI explora aprendizado contínuo com 8 GB de VRAM