Prompt engineering (or rather "Flow engineering") intensifies for code generation. Great reading and a reminder of how much alpha there is (pass@5 19% to 44%) in moving from a naive prompt:answer paradigm to a "flow" paradigm, where the answer is constructed iteratively.
The paper proposes qodo Flow, a code-oriented iterative flow that improves LLMs on code generation.
Besides achieving SoTA on a complex code generation dataset, I think the ideas and proposed methodology in this work are a big deal.
There's a new open-source, state-of-the-art code generation tool. It's a new approach that improves the performance of Large Language Models generating code.
The paper's authors call the process "qodo Flow" and tested it on the CodeContests dataset, which contains around 10,000 competitive programming problems.
The results put qodo Flow as the best approach to generate code we've seen. It beats DeepMind's AlphaCode and their new AlphaCode2 without needing to fine-tune a model!
qodo Flow is a test-based, multi-stage, code-oriented iterative flow, that improves the performances of LLMs on code problems. Tests on CodeContests, a difficult code-gen benchmark, show that accuracy improved GPT-4's accuracy from 19 - 44%.