Instella: Fully Open Language Models with Stellar Performance

TL;DR

Instella introduces a fully open, three-billion-parameter language model family that achieves state-of-the-art performance among open models, emphasizing transparency, reproducibility, and versatility across tasks.

Contribution

The paper presents Instella, a fully open-source LLM family trained on openly available data, with specialized variants for long context and mathematical reasoning, setting new benchmarks for open models.

Findings

Achieves state-of-the-art results among fully open models.

Competitiveness with leading open-weight models of similar size.

Demonstrates versatility with specialized variants for long contexts and math tasks.

Abstract

Large language models (LLMs) have demonstrated remarkable performance across a wide range of tasks, yet the majority of high-performing models remain closed-source or partially open, limiting transparency and reproducibility. In this work, we introduce Instella, a family of fully open three billion parameter language models trained entirely on openly available data and codebase. Powered by AMD Instinct MI300X GPUs, Instella is developed through large-scale pre-training, general-purpose instruction tuning, and alignment with human preferences. Despite using substantially fewer pre-training tokens than many contemporaries, Instella achieves state-of-the-art results among fully open models and is competitive with leading open-weight models of comparable size. We further release two specialized variants: Instella-Long, capable of handling context lengths up to 128K tokens, and Instella-Math, a reasoning-focused model enhanced through supervised fine-tuning and reinforcement learning on mathematical tasks. Together, these contributions establish Instella as a transparent, performant, and versatile alternative for the community, advancing the goal of open and reproducible language modeling research.