Apriel-Nemotron-15B-Thinker

TL;DR

Apriel-Nemotron-15B-Thinker is a compact 15-billion parameter language model that achieves competitive performance with larger models while significantly reducing memory and computational requirements.

Contribution

Introduces a new 15-billion parameter LLM with a four-stage training pipeline that matches larger models' performance at half the size.

Findings

Performs on par or better than 32B models across benchmarks.

Reduces memory footprint by 50% compared to similar-sized models.

Maintains high reasoning capabilities with fewer resources.

Abstract

While large language models (LLMs) have achieved remarkable reasoning capabilities across domains like code, math and other enterprise tasks, their significant memory and computational costs often preclude their use in practical enterprise settings. To this end, we introduce Apriel-Nemotron-15B-Thinker, a 15-billion parameter model in the ServiceNow Apriel SLM series that achieves performance against medium sized state-of-the-art models such as o1-mini, QWQ32B, and EXAONE-Deep-32B while maintaining only half the memory footprint of those alternatives. Apriel-Nemotron-15B-Thinker model is trained in a four stage training pipeline including 1) Base Model upscaling, 2) Continual Pre-training 3) Supervised Fine-tuning (SFT) and 4) Reinforcement Learning using GRPO. Comprehensive evaluations across a diverse suite of benchmarks consistently demonstrate that our Apriel-Nemotron-15B-Thinker model matches or exceeds the performance of its 32-billion parameter counterparts, despite being less than half their size.