Salesforce/E1-Math-1.5B
Salesforce/E1-Math-1.5B is a 1.5 billion parameter language model fine-tuned from DeepSeek-R1-Distilled-Qwen-1.5B. Developed by Salesforce, it is specifically trained for Elastic Reasoning, enabling adaptive problem-solving even with limited computational budget. This model excels at mathematical tasks by generalizing effectively to unseen budget constraints without additional training, making it suitable for resource-constrained reasoning applications.
Loading preview...
Overview
Salesforce/E1-Math-1.5B is a 1.5 billion parameter language model developed by Salesforce, fine-tuned from DeepSeek-R1-Distilled-Qwen-1.5B. Its core innovation lies in its training for Elastic Reasoning, a strategy that allows the model to adapt its reasoning process when computational resources are limited. This approach, integrated into GRPO, teaches the model to perform effectively even when its thinking process is cut short, and it generalizes well to varying budget constraints without requiring further training.
Key Capabilities
- Elastic Reasoning: Adapts its reasoning depth based on available computational budget.
- Mathematical Task Performance: Shows proficiency in mathematical problem-solving, as indicated by its performance metrics.
- Resource-Efficient Reasoning: Designed to generalize effectively to unseen budget constraints, making it suitable for scenarios where computational resources are variable or limited.
Performance Highlights
The model's performance is evaluated using an "Avg@16" metric, comparing it against DeepScaleR-1.5B. E1-Math-1.5B demonstrates competitive accuracy across different token budgets, showcasing its ability to reason adaptively. For instance, at a token budget of 1340, it achieves 13.5% accuracy, and at 3377 tokens, it reaches 27.9% accuracy.
Use Cases
- Budget-Constrained Reasoning: Ideal for applications requiring intelligent reasoning under strict computational or time limits.
- Mathematical Problem Solving: Suitable for tasks that benefit from adaptive and efficient mathematical reasoning.
- Research in Adaptive AI: A valuable tool for researchers exploring scalable chain-of-thought methods and elastic reasoning paradigms, as detailed in the associated paper "Scalable Chain of Thoughts via Elastic Reasoning" (arXiv:2505.05315).
Ethical Considerations
This model is released for research purposes only in support of an academic paper. Users are strongly advised to evaluate and address potential concerns regarding accuracy, safety, and fairness before deployment, especially in high-risk scenarios. It is crucial to consider common AI limitations and comply with applicable laws.