README.md

---
library_name: transformers
datasets:
- agentica-org/DeepScaleR-Preview-Dataset
- agentica-org/DeepCoder-Preview-Dataset
base_model:
- nvidia/AceReason-Nemotron-14B
license: cc-by-nc-4.0
---

## Introduction
E1-AceReason-14B is a language model fine-tuned from AceReason-Nemotron-14B. It is trained for Elastic Reasoning by budget-constrained rollout strategy, integrated into GRPO, which teaches the model to reason adaptively when the thinking process is cut short and generalizes effectively to unseen budget constraints without additional training.

## Usage
For detailed usage, please refer to [repo](https://github.com/SalesforceAIResearch/Elastic-Reasoning).

## Performance on AIME24 (Avg@16)

Note: We did not tune the training hyperparameters. The performance may slightly differ from the results reported in the original paper due to differences in the environment setup.

| Model | Tokens | Acc (%) | Tokens | Acc (%) | Tokens | Acc (%) | Tokens | Acc (%) | Tokens | Acc (%) |
|---------------|--------------|---------------|--------------|---------------|--------------|---------------|--------------|---------------|--------------|---------------|
| AceReason-Nemotron-14B | 13833 | 76.5 | - | -  | - | - | - | - | - | - |
| E1-AceReason-14B  | 8376 | 75.4 | 1318 | 13.3 | 1736 | 22.7 | 2660  | 33.8 | 3448 | 44.6 |

## Performance on LiveCodeBenchv5 (Avg@8)

Note: We did not tune the training hyperparameters. The performance may slightly differ from the results reported in the original paper due to differences in the environment setup.

| Model | Tokens | Acc (%) | Tokens | Acc (%) | Tokens | Acc (%) | Tokens | Acc (%) | Tokens | Acc (%) |
|---------------|--------------|---------------|--------------|---------------|--------------|---------------|--------------|---------------|--------------|---------------|
| AceReason-Nemotron-14B | 16669 | 59.2 | - | -  | - | - | - | - | - | - |
| E1-AceReason-14B  | 10796 | 57.8 | 1314 | 31.3 | 1810 | 36.3 | 2743  | 40.2 | 3585 | 43.5 |


## Citation


```bibtex
@article{xu2025scalable,
  title={Scalable Chain of Thoughts via Elastic Reasoning},
  author={Xu, Yuhui and Dong, Hanze and Wang, Lei and Sahoo, Doyen and Li, Junnan and Xiong, Caiming},
  journal={arXiv preprint arXiv:2505.05315},
  year={2025}
}
```

## Ethical Considerations
This release is for research purposes only in support of an academic paper. Our models, datasets, and code are not specifically designed or evaluated for all downstream purposes. We strongly recommend users evaluate and address potential concerns related to accuracy, safety, and fairness before deploying this model. We encourage users to consider the common limitations of AI, comply with applicable laws, and leverage best practices when selecting use cases, particularly for high-risk scenarios where errors or misuse could significantly impact people's lives, rights, or safety. For further guidance on use cases, refer to our AUP and AI AUP.