SliceLine ResearchTry Zeon

SliceLine Research

Research.

Technical reports from SliceLine model development.

Publications

Research paperAugust 202620.06M parameters · 200.0M tokens · 57.53 GPU-active min

Horizon: Architecture Screening and Scaling of a 20M-Parameter Language Model Under a One-Hour Consumer-GPU Budget

A wall-clock-first study of recurrent weight sharing, factorized lexical embeddings, external memory, and dense scaling

Technical reportAugust 2026Antares · 108.55M parameters

Antares: Resource-Constrained Pretraining and Post-Training Analysis of a 108M-Parameter Language Model

Resource-constrained scratch pretraining, post-training, runtime attribution, and zero-shot evaluation.

Technical reportAugust 2026Zeon-1 and Zeon-2 · ~54.8M parameters

Zeon: Development and Deployment of 54M-Parameter Language Models with Tool-Augmented Post-Training

A reconstruction of Zeon-1 and Zeon-2 from training notebooks, deployment artifacts, and contemporaneous launch records.

Technical reportAugust 2026Miranda family · 2024

MirandaGPT: Iterative Conversational Language-Model Development in 2024

A post-hoc technical analysis of the Miranda and MirandaGPT artifacts preceding the Zeon family.

Technical reportAugust 2026MICRO series · 2024

The MICRO Models: Early Recurrent Language Modeling on Conversational Data

A technical reconstruction of MICRO-1, MICRO-3, and related 2024 experiments.

Technical reportAugust 2026Selected experiments · 2023–2024

Foundations: Early Neural and Transformer Language-Model Experiments, 2023-2024

Selected artifacts preceding the MICRO, Miranda, Zeon, and Antares model families.

Technical reportAugust 20262024–2026 model lineage

Two Years of Small Language-Model Development: From Early Recurrent Models to Antares

A two-year retrospective on small language-model development under limited compute.