Overview

AbleCredit-R0-Qwen-2.5-3B-Instruct is a 3.1 billion parameter instruction-tuned language model developed by AbleCredit (LightBees Technologies Private Limited). It is built upon the Qwen 2.5 3B Instruct base model and has been fine-tuned using Deepseek R1 style (GRPO) reinforcement learning with rule-based rewards.

Key Capabilities & Training

• Reinforcement Learning: Utilizes GRPO (Deepseek R1 style) for fine-tuning, focusing on enhancing reasoning abilities.
• Specialized Training Data: Trained on a combination of open-source logical reasoning datasets and a proprietary finance dataset created by AbleCredit.com.
• Reasoning Focus: Designed with a primary intent for research in applying small LLMs to financial applications, particularly credit underwriting.

Performance

• GSM8K Benchmark: Achieves approximately 67% score on the GSM8K mathematical reasoning benchmark in a zero-shot setting, indicating strong logical reasoning without specific examples.

Usage & Licensing

• Hugging Face Integration: Compatible with standard Hugging Face setups for easy deployment and interaction.
• License: Retains the original Qwen research license, which does not permit commercial use.