GreenMind-Medium-14B-R1: A Vietnamese Reasoning LLM
GreenMind-Medium-14B-R1 is a 14.7 billion parameter causal language model developed by GreenNode, fine-tuned from Qwen/Qwen2.5-14B-Instruct. It is specifically designed for the Vietnamese language and focuses on addressing questions requiring intermediate-level reasoning.
Key Capabilities & Differentiators
- Enhanced Reasoning: Utilizes the Group Relative Policy Optimization strategy to produce logically coherent responses for complex questions in areas like mathematics, natural science, and social science.
- Vietnamese Language Specialization: Developed to excel in Vietnamese, outperforming its base model and other larger models on various Vietnamese benchmarks.
- Extended Context Window: Supports a full context length of 131,072 tokens, allowing for processing extensive inputs.
Performance Highlights
Evaluations demonstrate GreenMind-Medium-14B-R1's strong performance:
- SeaExam Dataset: Achieved an average score of 72.79, surpassing Meta-Llama-3.1-70B-Instruct (69.7) and its base model Qwen2.5-14B-Instruct (69.8) on Vietnamese-specific reasoning tasks.
- VLSP 2023 Challenge: Significantly outperformed prior GreenNode models and other SOTA models across multiple Vietnamese NLP tasks, including ComprehensionQA-vi (0.8689), Exams-vi (0.7796), and MMLU-vi (0.7124).
Ideal Use Cases
This model is particularly well-suited for applications requiring:
- Complex Question Answering in Vietnamese: Excels at problems demanding step-by-step logical deduction.
- Educational Tools: Can be used for generating explanations and solutions for academic subjects in Vietnamese.
- Content Generation: Capable of producing coherent and reasoned text in Vietnamese for various domains.