Reasoning Under 1 Billion: Memory-Augmented Reinforcement Learning for Large Language Models
Fuente:
arXiv
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912306863013888 |
|---|---|
| author | Le, Hung Do, Dai Nguyen, Dung Venkatesh, Svetha |
| author_facet | Le, Hung Do, Dai Nguyen, Dung Venkatesh, Svetha |
| contents | Recent advances in fine-tuning large language models (LLMs) with reinforcement learning (RL) have shown promising improvements in complex reasoning tasks, particularly when paired with chain-of-thought (CoT) prompting. However, these successes have been largely demonstrated on large-scale models with billions of parameters, where a strong pretraining foundation ensures effective initial exploration. In contrast, RL remains challenging for tiny LLMs with 1 billion parameters or fewer because they lack the necessary pretraining strength to explore effectively, often leading to suboptimal reasoning patterns. This work introduces a novel intrinsic motivation approach that leverages episodic memory to address this challenge, improving tiny LLMs in CoT reasoning tasks. Inspired by human memory-driven learning, our method leverages successful reasoning patterns stored in memory while allowing for controlled exploration to generate novel responses. Intrinsic rewards are computed efficiently using a kNN-based episodic memory, allowing the model to discover new reasoning strategies while quickly adapting to effective past solutions. Experiments on fine-tuning GSM8K and AI-MO datasets demonstrate that our approach significantly enhances smaller LLMs' sample efficiency and generalization capability, making RL-based reasoning improvements more accessible in low-resource settings. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2504_02273 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Reasoning Under 1 Billion: Memory-Augmented Reinforcement Learning for Large Language Models Le, Hung Do, Dai Nguyen, Dung Venkatesh, Svetha Machine Learning Recent advances in fine-tuning large language models (LLMs) with reinforcement learning (RL) have shown promising improvements in complex reasoning tasks, particularly when paired with chain-of-thought (CoT) prompting. However, these successes have been largely demonstrated on large-scale models with billions of parameters, where a strong pretraining foundation ensures effective initial exploration. In contrast, RL remains challenging for tiny LLMs with 1 billion parameters or fewer because they lack the necessary pretraining strength to explore effectively, often leading to suboptimal reasoning patterns. This work introduces a novel intrinsic motivation approach that leverages episodic memory to address this challenge, improving tiny LLMs in CoT reasoning tasks. Inspired by human memory-driven learning, our method leverages successful reasoning patterns stored in memory while allowing for controlled exploration to generate novel responses. Intrinsic rewards are computed efficiently using a kNN-based episodic memory, allowing the model to discover new reasoning strategies while quickly adapting to effective past solutions. Experiments on fine-tuning GSM8K and AI-MO datasets demonstrate that our approach significantly enhances smaller LLMs' sample efficiency and generalization capability, making RL-based reasoning improvements more accessible in low-resource settings. |
| title | Reasoning Under 1 Billion: Memory-Augmented Reinforcement Learning for Large Language Models |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2504.02273 |