# Mathematical Reasoning Enhancement

UniReason-Qwen3-14B-RL is a quantized model applicable to multiple domains, especially proficient in text generation and mathematical reasoning tasks.

A fine-tuned version based on Qwen3-1.7B, which enhances mathematical reasoning ability through one-shot reinforcement learning and verifiable reward (RLVR) methods, and performs excellently in mathematical benchmark tests and coding tasks.

A math and code reasoning model trained through reinforcement learning, based on DeepSeek-R1-Distilled-Qwen-7B, excelling in mathematical and code reasoning tasks

A Japanese dialogue model optimized through merging Qwen2.5-32B and QwQ-32B, enhanced with Chat Vector and ORPO technologies for improved instruction following

ThinkEdit is a lightweight weight editing method that identifies and modifies a small number of attention heads to alleviate the issue of overly brief reasoning chains generated by inference models, thereby improving reasoning accuracy.

The first open-source implementation of large-scale reasoning-oriented reinforcement learning focusing on scalability, simplicity, and ease of use

Granite-8B-Code-Instruct-4K is an 8-billion-parameter code instruction model, fine-tuned on various permissible instruction datasets based on Granite-8B-Code-Base-4K, enhancing its ability to follow instructions, including logical reasoning and problem-solving skills.

Granite-3B-Code-Instruct-2K is a 3-billion-parameter model fine-tuned from Granite-3B-Code-Base-2K, with enhanced instruction-following capabilities, particularly excelling in code generation and logical reasoning tasks.

MathGenie is a model that enhances the mathematical reasoning capabilities of large language models by generating synthetic data through question back-translation.

Enhances the reasoning generalization capabilities of large language models through reinforcement fine-tuning, based on Codellama fine-tuning, suitable for code generation and comprehension tasks.

A math problem-solving model based on Math-Shepherd's step-by-step reinforcement learning, excelling on GSM8K and MATH datasets