qihoo360/Light-R1-7B-DS
TEXT GENERATIONConcurrency Cost:1Model Size:7.6BQuant:FP8Ctx Length:32kPublished:Mar 12, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Warm
The Light-R1-7B-DS is a 7.6 billion parameter language model developed by Qihoo360, fine-tuned from DeepSeek-R1-Distill-Qwen-7B. It is specifically optimized for mathematical reasoning, achieving state-of-the-art performance among open-source 7B math models on AIME24 and AIME25 benchmarks. This model demonstrates strong mathematical problem-solving capabilities, even performing well on GPQA without specific training.
Loading preview...
Light-R1-7B-DS: State-of-the-Art 7B Math Model
Light-R1-7B-DS, developed by Qihoo360, is a 7.6 billion parameter model derived from DeepSeek-R1-Distill-Qwen-7B. It stands out as a leading open-source 7B math model, achieving impressive scores of 59.1 on AIME24 and 44.3 on AIME25, as detailed in its technical report.
Key Capabilities and Features
- Exceptional Math Performance: Achieves top-tier results on challenging mathematical benchmarks like AIME24 and AIME25.
- Efficient Training: Further trained with only 3,000 SFT data points, demonstrating the effectiveness of its released dataset.
- Broad Applicability: Performs well on general question-answering tasks (GPQA) despite its specialized math training.
- Data Decontamination: Underwent rigorous data decontamination processes, including exact and N-gram matching, to ensure benchmark integrity and prevent contamination from test sets.
When to Use This Model
- Mathematical Problem Solving: Ideal for applications requiring advanced mathematical reasoning and problem-solving.
- Research and Development: Suitable for researchers exploring efficient fine-tuning methods with limited data, given its strong performance from only 3K SFT data.
- Benchmarking: A strong candidate for evaluating and comparing mathematical capabilities against other 7B models.