AI Paper Atlas

Research map / Large language models / Agents, code and math

Coder Reviewer Reranking for Code Generation

Tianyi Zhang, Tao Yu, Tatsunori Hashimoto, Mike Lewis, Wen-tau Yih, Daniel Fried, Sida Wang

ICML 2023 · Large language models › Agents, code and math · approach: Execution-based code verification · 7 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Add Reviewer likelihood to Coder-only rerankingCode generation reranking across six datasets and model families · 48 dataset-model pairsaccuracy: 43 out of 48Improved
Use length-normalized Coder-Reviewer scores instead of unnormalized scores (ablation)Coder-Reviewer · HumanEval, Codex002accuracy: 56.7 -> 62.5Improved
Use the alternate MMI formulation based on subtracting the marginal program likelihood (ablation)Coder-Reviewer · HumanEvalaccuracy: 56.7 -> 55.30; 62.5 -> 50.73Hurt
Use the alternate MMI formulation based on subtracting the marginal program likelihood (ablation)Coder-Reviewer · MBPP-Saccuracy: 64.4 -> 63.0; 61.6 -> 62.4Hurt
Tune the Coder/Reviewer mixing ratio instead of fixing it at 0.5 (ablation)Coder-Reviewer · Most experimental settingsaccuracy: less than 1%Improved
Tune the Coder/Reviewer mixing ratio (ablation)InCoder6B on Plotting and CodeGen16B on NL2Bash · Plotting; NL2BashaccuracyImproved
Remove degenerate-solution rejection (ablation)Codex002 reranking methods · HumanEvalaccuracy: 61.5 -> 53.2Hurt
Remove degenerate-solution rejection (ablation)Codex002 reranking methods · MBPP-Saccuracy: 60.8 -> 60.5Hurt
Filter empty, trivial, repetitive, and surface-overlap-prone programs before reranking (ablation)Coder, Reviewer, and Coder-Reviewer · HumanEval and MBPP-Saccuracy: 38.1 with 7.1 degradation; 55.3 with 4.5 degradationImproved
Apply executability filtering before reranking (ablation)Most ranking methods · Most datasetsaccuracyImproved
Increase the number of sampled programs reranked (ablation)Coder-Reviewer versus Coder-only ranking · Zero-shot datasetsaccuracyMixed
Use Reviewer-only ranking instead of Coder-only rankingCodeGen models · Across datasetsaccuracy: 5 exceptions out of 48Mixed

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy