OrionLLM/GRM-1.5b: A Reasoning-Focused 1.5B Model

GRM-1.5b is a compact yet powerful 1.5 billion parameter model developed by OrionLLM, specifically engineered for general-purpose reasoning. It excels in multi-domain problem-solving, encompassing areas like mathematics, logic, coding, and medical reasoning. This model is optimized for dedicated reasoning behavior, promoting stepwise problem-solving and improved consistency in its outputs.

Key Capabilities

• Enhanced Multi-Domain Reasoning: Fine-tuned across a diverse mixture of reasoning, code, math, and medical reasoning data.
• Efficiency: At 1.5B parameters, it is small and efficient, making it practical for local inference and experimentation.
• Fine-Tune Friendly: Designed as an excellent starting point for further supervised fine-tuning (SFT), GRPO, or DPO pipelines.
• Strong Performance: Demonstrates competitive performance against other 1.5B models in various reasoning benchmarks, notably achieving 52.0 on AIME24, 87.0 on AMC23, and 27.3 on HMMT O2/25.

Good For

• Developers seeking a lightweight, general-purpose model for daily reasoning tasks.
• Applications requiring robust performance in math, logic, and coding challenges.
• As a foundational model for custom fine-tuning to specific reasoning-intensive domains.
• Local deployment and experimentation due to its efficient size.