57 lines
2.7 KiB
Markdown
57 lines
2.7 KiB
Markdown
|
---
|
|||
|
|
base_model:
|
|||
|
|
- deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
|
|||
|
|
datasets:
|
|||
|
|
- knoveleng/open-rs
|
|||
|
|
- knoveleng/open-s1
|
|||
|
|
- knoveleng/open-deepscaler
|
|||
|
|
license: mit
|
|||
|
|
pipeline_tag: text-generation
|
|||
|
|
inference: true
|
|||
|
|
library_name: transformers
|
|||
|
|
---
|
|||
|
|
|
|||
|
|
# Model Summary
|
|||
|
|
|
|||
|
|
This repository hosts model for the **Open RS** project, accompanying the paper *Reinforcement Learning for Reasoning in Small LLMs: What Works and What Doesn’t*. The project explores enhancing reasoning capabilities in small large language models (LLMs) using reinforcement learning (RL) under resource-constrained conditions.
|
|||
|
|
|
|||
|
|
We focus on a 1.5-billion-parameter model, `DeepSeek-R1-Distill-Qwen-1.5B`, trained on 4 NVIDIA A40 GPUs (48 GB VRAM each) within 24 hours. By adapting the Group Relative Policy Optimization (GRPO) algorithm and leveraging a curated, compact mathematical reasoning dataset, we conducted three experiments to assess performance and behavior. Key findings include:
|
|||
|
|
|
|||
|
|
- Significant reasoning improvements, e.g., AMC23 accuracy rising from 63% to 80% and AIME24 reaching 46.7%, outperforming `o1-preview`.
|
|||
|
|
- Efficient training with just 7,000 samples at a cost of $42, compared to thousands of dollars for baseline models.
|
|||
|
|
- Challenges like optimization instability and length constraints with extended training.
|
|||
|
|
|
|||
|
|
These results showcase RL-based fine-tuning as a cost-effective approach for small LLMs, making reasoning capabilities accessible in resource-limited settings. We open-source our code, models, and datasets to support further research.
|
|||
|
|
|
|||
|
|
For more details, please refer our [github](https://github.com/knoveleng/open-rs).
|
|||
|
|
|
|||
|
|
## Evaluation
|
|||
|
|
### Performance Highlights
|
|||
|
|
- **Open-RS1**: 53.0% avg. score
|
|||
|
|
- **Open-RS2**: 55.7% avg. score, 80.0% on AMC23
|
|||
|
|
- **Open-RS3**: 56.3% avg. score, 46.7% on AIME24 (outperforms `o1-preview` at 44.6%)
|
|||
|
|
- Competitive MATH-500 scores; Minerva lags behind 7B models.
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
### Cost Efficiency
|
|||
|
|
Our approach uses 7,000 samples (42,000 total outputs) and costs ~$42 on 4x A40 GPUs in 24 hours, compared to:
|
|||
|
|
- 7B models: `Qwen2.5-7B-SimpleRL` ($1,633), `Eurus-2-7B-PRIME` ($1,088)
|
|||
|
|
- 1.5B models: `DeepScaleR-1.5B-Preview` ($3,629), `Still-3-1.5B-Preview` ($2,268)
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
## Citation
|
|||
|
|
If this project aids your work, please cite it as:
|
|||
|
|
```
|
|||
|
|
@inproceedings{
|
|||
|
|
dang2026reinforcement,
|
|||
|
|
title={Reinforcement Learning for Reasoning in Small {LLM}s: What Works and What Doesn{\textquoteright}t},
|
|||
|
|
author={Quy-Anh Dang and Chris Ngo},
|
|||
|
|
booktitle={Logical and Symbolic Reasoning in Language Models @ AAAI 2026},
|
|||
|
|
year={2026},
|
|||
|
|
url={https://openreview.net/forum?id=3pWL6Zxc4A}
|
|||
|
|
}
|
|||
|
|
```
|