Update README.md

README.md

@@ -36,7 +36,7 @@ For full transparency and reproducibility, please refer to our technical report
 
 
 
-🚀 The **JT-Math-8B-Instruct** is an 8-billion parameter language model built on the **Jiutian LLM architecture** with a **context length of 32,768 tokens**. Its development involved two key stages: initial pre-training of the **JT-Math-8B-Base** model on a diverse corpus of text and mathematical data, followed by a two-stage instruction tuning process. This tuning began with **Supervised Fine-Tuning (SFT)**, where the model was trained on a high-quality, multilingual dataset of mathematical problems and solutions in both English and Chinese to grasp problem-solving patterns. Subsequently, **Reinforcement Learning (RL)** was applied within an 8K context window to enhance reasoning accuracy, minimize logical fallacies, and align the model more closely with human preferences for clear and correct mathematical solutions.
+🚀 The **JT-Math-8B-Instruct** is an 8-billion parameter language model built on the **Jiutian LLM architecture** with a **context length of 32,768 tokens**. Its development involved two key stages: initial pre-training of the **JT-Math-8B-Base** model on a diverse corpus of text and mathematical data, followed by a two-stage instruction tuning process. This tuning began with **Supervised Fine-Tuning (SFT)**, where the model was trained on a high-quality, multilingual dataset of mathematical problems and solutions in both English and Chinese to grasp problem-solving patterns. Subsequently, **Reinforcement Learning (RL)** was applied to enhance reasoning accuracy, minimize logical fallacies, and align the model more closely with human preferences for clear and correct mathematical solutions.