Towards Efficient Fine-Tuning of Language Models With Organizational Data for Automated Software Review
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Large language models like BERT and GPT possess significant capabilities and potential impacts across various applications. Software engineers often use these models for code-related tasks, including generating, debugging, and summarizing code. Nevertheless, large language models still have several flaws, including model hallucination. (e.g., generating erroneous code and producing outdated and inaccurate programs) and the substantial computational resources and energy required for training and fine-tuning. To tackle these challenges, we propose CodeMentor, a framework for few-shot learning to train large language models with the data available within the organization. We employ the framework to train a language model for code review activities, such as code refinement and review generation. The framework utilizes heuristic rules and weak supervision techniques to leverage available data, such as previous review comments, issue reports, and related code updates. Then, the framework employs the constructed dataset to fine-tune LLMs for code review tasks. Additionally, the framework integrates domain expertise by employing reinforcement learning with human feedback. This allows domain experts to assess the generated code and enhance the model performance. Also, to assess the performance of the proposed model, we evaluate it with four state-of-the-art techniques in various code review tasks. The experimental results attest that CodeMentor enhances the performance in all tasks compared to the state-of-the-art approaches, with an improvement of up to 22.3%, 43.4%, and 24.3% in code quality estimation, review generation, and bug report summarization tasks, respectively.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.002 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it