Saved in:
Bibliographic Details
Main Author: Grigorev, George
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2602.02522
Tags: Add Tag
No Tags, Be the first to tag this record!
Table of Contents:
  • We present IMU-1, a 430M-parameter language model trained on 72B tokens that approaches the benchmark performance of models trained on 56x more data. We describe a validated training recipe combining recent architectural interventions (QK-norm attention, per-head gating, value residuals, LayerNorm scaling) with optimization advances (NorMuon with cautious weight decay, muP parametrization) and a three-stage training schedule with post-hoc checkpoint EMA. We provide ablations for each component and release code, weights and data to enable reproduction: https://huggingface.co/thepowerfuldeez/imu1_base