TIMO: Towards Better Temporal Reasoning for Language Models This paper presents TIMO, a model designed to enhance temporal reasoning capabilities in large language models (LLMs). The authors systematically study 38 temporal reasoning tasks and find that 19 are directly related to mathematics. They propose a self-critic temporal optimization method to improve the model's temporal reasoning abilities without sacrificing general task performance. TIMO outperforms other LLMs in average accuracy scores and achieves state-of-the-art performance on comparable sizes. The model is trained on a combination of mathematical and temporal reasoning tasks, and extensive experiments validate its effectiveness across diverse temporal tasks. The code is available at https://github.com/zhaochen0110/Timo. The paper also discusses the correlation between mathematics and temporal reasoning, and introduces a self-critic temporal task optimization method to enhance the model's temporal reasoning capabilities. The results show that TIMO significantly improves performance on both math-time and pure-time tasks, demonstrating its effectiveness in temporal reasoning. The model is trained on a large-scale dataset and shows strong generalization across different temporal reasoning scenarios. The paper also compares TIMO with other models, including MATHLLAMA and TIMELLAMA, and finds that TIMO outperforms them in most tasks. The authors conclude that their framework provides a comprehensive solution for temporal reasoning in LLMs.TIMO: Towards Better Temporal Reasoning for Language Models This paper presents TIMO, a model designed to enhance temporal reasoning capabilities in large language models (LLMs). The authors systematically study 38 temporal reasoning tasks and find that 19 are directly related to mathematics. They propose a self-critic temporal optimization method to improve the model's temporal reasoning abilities without sacrificing general task performance. TIMO outperforms other LLMs in average accuracy scores and achieves state-of-the-art performance on comparable sizes. The model is trained on a combination of mathematical and temporal reasoning tasks, and extensive experiments validate its effectiveness across diverse temporal tasks. The code is available at https://github.com/zhaochen0110/Timo. The paper also discusses the correlation between mathematics and temporal reasoning, and introduces a self-critic temporal task optimization method to enhance the model's temporal reasoning capabilities. The results show that TIMO significantly improves performance on both math-time and pure-time tasks, demonstrating its effectiveness in temporal reasoning. The model is trained on a large-scale dataset and shows strong generalization across different temporal reasoning scenarios. The paper also compares TIMO with other models, including MATHLLAMA and TIMELLAMA, and finds that TIMO outperforms them in most tasks. The authors conclude that their framework provides a comprehensive solution for temporal reasoning in LLMs.