A from-scratch study isolates three design axes of looped language models—recurrence placement, input injection, and recurrent state organization—by training models at matched parameter scale, logical depth, and token budget. The work argues that established looped architectures entangle these choices, and it separates them to test each independently. The findings are presented as Part I of a series on building looped models correctly.
Looped language models reuse physical Transformer blocks across logical depth, but established architectures entangle recurrence placement, input injection, and recurrent state organization. We isolate these axes in models trained from scratch at matched parameter scale, logical depth, and token bud
EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.