PyBench: Evaluating LLM Agent on various real-world coding tasks

TL;DR

PyBench is a comprehensive benchmark designed to evaluate large language model agents on diverse real-world Python coding tasks, highlighting current limitations and demonstrating the effectiveness of a fine-tuned 8B model.

Contribution

Introduces PyBench, a new benchmark covering various real-world coding tasks, and presents PyLlama3, a fine-tuned model that outperforms larger models on this benchmark.

Findings

Current open-source LLMs struggle with PyBench tasks.

PyLlama3 surpasses larger models in performance.

Comprehensive reasoning and feedback incorporation are essential.

Abstract

The LLM Agent, equipped with a code interpreter, is capable of automatically solving real-world coding tasks, such as data analysis and image editing. However, existing benchmarks primarily focus on either simplistic tasks, such as completing a few lines of code, or on extremely complex and specific tasks at the repository level, neither of which are representative of various daily coding tasks. To address this gap, we introduce \textbf{PyBench}, a benchmark encompassing five main categories of real-world tasks, covering more than 10 types of files. Given a high-level user query and related files, the LLM Agent needs to reason and execute Python code via a code interpreter for a few turns before making a formal response to fulfill the user's requirements. Successfully addressing tasks in PyBench demands a robust understanding of various Python packages, superior reasoning capabilities, and the ability to incorporate feedback from executed code. Our evaluations indicate that current open-source LLMs are struggling with these tasks. Hence, we conduct analysis and experiments on four kinds of datasets proving that comprehensive abilities are needed for PyBench. Our fine-tuned 8B size model: \textbf{PyLlama3} achieves an exciting performance on PyBench which surpasses many 33B and 70B size models. Our Benchmark, Training Dataset, and Model are available at: {https://github.com/Mercury7353/PyBench}