# GRPO Optimization

INTELLECT-2 is a large language model with 32 billion parameters released by Prime Intellect. It is built on the Qwen2 architecture and focuses on mathematical, coding, and logical reasoning tasks.

A fine-tuned version based on Qwen/Qwen3-8B using the medmcqa-grpo dataset, specialized in medical multiple-choice question answering tasks

E1-Math-1.5B is a language model fine-tuned based on DeepSeek-R1-Distilled-Qwen-1.5B, supporting elastic reasoning and the GRPO method, suitable for budget-constrained deduction scenarios.

XiYanSQL-QwenCoder-2504 is the latest SQL generation model, combining fine-tuning and GRPO training, supporting multiple dialects with efficient and accurate SQL generation capabilities.

A multimodal language model fine-tuned by Tencent PCG Basic Algorithm Center, optimized based on Qwen2.5-VL-3B-Instruct, achieving state-of-the-art performance in multiple multimodal reasoning benchmarks among models of the same scale

Yehia is an empathetic Arabic language model, developed based on ALLaM-7B-Instruct-preview, offering considerate, friendly, and practical Arabic and English conversations.

A reinforcement learning model based on Qwen2.5-3B-Instruct, specifically designed for conflict event classification, optimized using the GRPO method for multi-reward signals and structured reasoning formats.