Unlock the Correlation between Supervised Fine-Tuning and Reinforcement Learning in Training Code Large Language Models
June 14, 2024 Β· Declared Dead Β· π arXiv.org
"No code URL or promise found in abstract"
Evidence collected by the PWNC Scanner
Authors
Jie Chen, Xintian Han, Yu Ma, Xun Zhou, Liang Xiang
arXiv ID
2406.10305
Category
cs.SE: Software Engineering
Cross-listed
cs.AI,
cs.LG
Citations
3
Venue
arXiv.org
Last Checked
4 months ago
Abstract
Automatic code generation has been a longstanding research topic. With the advancement of general-purpose large language models (LLMs), the ability to code stands out as one important measure to the model's reasoning performance. Usually, a two-stage training paradigm is implemented to obtain a Code LLM, namely the pretraining and the fine-tuning. Within the fine-tuning, supervised fine-tuning (SFT), and reinforcement learning (RL) are often used to improve the model's zero-shot ability. A large number of work has been conducted to improve the model's performance on code-related benchmarks with either modifications to the algorithm or refinement of the dataset. However, we still lack a deep insight into the correlation between SFT and RL. For instance, what kind of dataset should be used to ensure generalization, or what if we abandon the SFT phase in fine-tuning. In this work, we make an attempt to understand the correlation between SFT and RL. To facilitate our research, we manually craft 100 basis python functions, called atomic functions, and then a synthesizing pipeline is deployed to create a large number of synthetic functions on top of the atomic ones. In this manner, we ensure that the train and test sets remain distinct, preventing data contamination. Through comprehensive ablation study, we find: (1) Both atomic and synthetic functions are indispensable for SFT's generalization, and only a handful of synthetic functions are adequate; (2) Through RL, the SFT's generalization to target domain can be greatly enhanced, even with the same training prompts; (3) Training RL from scratch can alleviate the over-fitting issue introduced in the SFT phase.
Community Contributions
Found the code? Know the venue? Think something is wrong? Let us know!
π Similar Papers
In the same crypt β Software Engineering
R.I.P.
π»
Ghosted
R.I.P.
π»
Ghosted
Microservices: yesterday, today, and tomorrow
π
π
The Cartographer
A Survey of Machine Learning for Big Code and Naturalness
R.I.P.
π»
Ghosted
An Overview on Smart Contracts: Challenges, Advances and Platforms
R.I.P.
π»
Ghosted
Slither: A Static Analysis Framework For Smart Contracts
R.I.P.
π»
Ghosted
ContractFuzzer: Fuzzing Smart Contracts for Vulnerability Detection
Died the same way β π» Ghosted
R.I.P.
π»
Ghosted
Federated Learning: Strategies for Improving Communication Efficiency
R.I.P.
π»
Ghosted
In-Datacenter Performance Analysis of a Tensor Processing Unit
R.I.P.
π»
Ghosted
Deep Convolutional Neural Networks for Computer-Aided Detection: CNN Architectures, Dataset Characteristics and Transfer Learning
R.I.P.
π»
Ghosted