Bohdan Petryshyn, Mantas Lukoševičius
Recent advancements in Large Language Models (LLMs) and their utilization in code generation tasks have significantly reshaped the field of software development. Despite the remarkable efficacy of code completion solutions in mainstream programming languages, their performance lags when applied to less ubiquitous formats such as OpenAPI definitions. This study evaluates the OpenAPI completion performance of GitHub Copilot, a prevalent commercial code completion tool, and proposes a set of task-specific optimizations leveraging Meta's open-source model Code Llama. A semantics-aware OpenAPI completion benchmark proposed in this research is used to perform a series of experiments through which the impact of various prompt-engineering and fine-tuning techniques on the Code Llama model's performance is analyzed. The fine-tuned Code Llama model reaches a peak correctness improvement of 55.2% over GitHub Copilot despite utilizing 25 times fewer parameters than the commercial solution's underlying Codex model. Additionally, this research proposes an enhancement to a widely used code infilling training technique, addressing the issue of underperformance when the model is prompted with context sizes smaller than those used during training. The dataset, the benchmark, and the model fine-tuning code are made publicly available.
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Code Completion | OpenAPI completion refined | Correctness, avg., % | 34 | Code Llama 7B, fine-tuned with document splitting |
| Code Completion | OpenAPI completion refined | Correctness, max., % | 42 | Code Llama 7B, fine-tuned with document splitting |
| Code Completion | OpenAPI completion refined | Validness, avg., % | 69.1 | Code Llama 7B, fine-tuned with document splitting |
| Code Completion | OpenAPI completion refined | Validness, max., % | 76 | Code Llama 7B, fine-tuned with document splitting |
| Code Completion | OpenAPI completion refined | Correctness, avg., % | 32 | Code Llama 7B, fine-tuned at 4096 tokens |
| Code Completion | OpenAPI completion refined | Correctness, max., % | 45 | Code Llama 7B, fine-tuned at 4096 tokens |
| Code Completion | OpenAPI completion refined | Validness, avg., % | 63.1 | Code Llama 7B, fine-tuned at 4096 tokens |
| Code Completion | OpenAPI completion refined | Validness, max., % | 84 | Code Llama 7B, fine-tuned at 4096 tokens |
| Code Completion | OpenAPI completion refined | Correctness, avg., % | 31.1 | Code Llama 7B |
| Code Completion | OpenAPI completion refined | Correctness, max., % | 36 | Code Llama 7B |
| Code Completion | OpenAPI completion refined | Validness, avg., % | 60.7 | Code Llama 7B |
| Code Completion | OpenAPI completion refined | Validness, max., % | 64 | Code Llama 7B |
| Code Completion | OpenAPI completion refined | Correctness, avg., % | 29 | GitHub Copilot |
| Code Completion | OpenAPI completion refined | Correctness, max., % | 29 | GitHub Copilot |
| Code Completion | OpenAPI completion refined | Validness, avg., % | 68 | GitHub Copilot |
| Code Completion | OpenAPI completion refined | Validness, max., % | 68 | GitHub Copilot |
| OpenAPI code completion | OpenAPI completion refined | Correctness, avg., % | 34 | Code Llama 7B, fine-tuned with document splitting |
| OpenAPI code completion | OpenAPI completion refined | Correctness, max., % | 42 | Code Llama 7B, fine-tuned with document splitting |
| OpenAPI code completion | OpenAPI completion refined | Validness, avg., % | 69.1 | Code Llama 7B, fine-tuned with document splitting |
| OpenAPI code completion | OpenAPI completion refined | Validness, max., % | 76 | Code Llama 7B, fine-tuned with document splitting |
| OpenAPI code completion | OpenAPI completion refined | Correctness, avg., % | 32 | Code Llama 7B, fine-tuned at 4096 tokens |
| OpenAPI code completion | OpenAPI completion refined | Correctness, max., % | 45 | Code Llama 7B, fine-tuned at 4096 tokens |
| OpenAPI code completion | OpenAPI completion refined | Validness, avg., % | 63.1 | Code Llama 7B, fine-tuned at 4096 tokens |
| OpenAPI code completion | OpenAPI completion refined | Validness, max., % | 84 | Code Llama 7B, fine-tuned at 4096 tokens |
| OpenAPI code completion | OpenAPI completion refined | Correctness, avg., % | 31.1 | Code Llama 7B |
| OpenAPI code completion | OpenAPI completion refined | Correctness, max., % | 36 | Code Llama 7B |
| OpenAPI code completion | OpenAPI completion refined | Validness, avg., % | 60.7 | Code Llama 7B |
| OpenAPI code completion | OpenAPI completion refined | Validness, max., % | 64 | Code Llama 7B |
| OpenAPI code completion | OpenAPI completion refined | Correctness, avg., % | 29 | GitHub Copilot |
| OpenAPI code completion | OpenAPI completion refined | Correctness, max., % | 29 | GitHub Copilot |
| OpenAPI code completion | OpenAPI completion refined | Validness, avg., % | 68 | GitHub Copilot |
| OpenAPI code completion | OpenAPI completion refined | Validness, max., % | 68 | GitHub Copilot |