Research map / Large language models / Agents, code and math
CodeGen: An Open Large Language Model for Code with Multi-Turn Program Synthesis
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 changed | Where | Measured | Effect |
|---|---|---|---|
| Factorize the user specification into multiple turns instead of concatenating it into one prompt (ablation) | CodeGen-Mono · MTPB | pass rate: 25.43 -> 38.72; 28.48 -> 43.52; 38.74 -> 47.34 | Helped |
| Increase model size from 350M to 16.1B | CodeGen-Mono · MTPB | pass rate: 16.98 -> 47.34 | Helped |
| Continue training on multilingual code from BigQuery (ablation) | CodeGen · HumanEval | pass@1: 10.43 -> 18.16 | Helped |
| Continue training on Python-only BigPython data (ablation) | CodeGen-Multi · HumanEval | pass@1: 18.16 -> 26.13 | Helped |
| Continue training on Python-only BigPython data (ablation) | CodeGen-Multi · MTPB | pass rate: 20.82 -> 38.72 | Helped |
| Use multi-turn factorization on easy problems with larger models (ablation) | CodeGen-Mono 6.1B and 16.1B · MTPB easy problems | pass-rate improvement: 0.19%; -0.25% | No clear effect |
| Use multi-turn prompts rather than concatenated single-turn prompts (ablation) | CodeGen-Mono 350M · MTPB | pass rate: 5.75 -> 16.98 | Helped |
| Increase CodeGen-Mono model size from 350M to 16.1B | CodeGen-Mono · HumanEval | pass@1: 12.76 -> 29.28 | Helped |
Closest papers by meaning
- CodeChemist: Test-Time Scaling for Low-Resource Code Generation via Functional Knowledge Transfer (ICML 2026 · similarity 0.90)
- CodeChain: Towards Modular Code Generation Through Chain of Self-revisions with Representative Sub-modules (ICLR 2024 · similarity 0.90)
- ConvCodeWorld: Benchmarking Conversational Code Generation in Reproducible Feedback Environments (ICLR 2025 · similarity 0.90)
- BigCodeBench: Benchmarking Code Generation with Diverse Function Calls and Complex Instructions (ICLR 2025 · similarity 0.90)
- How efficient is LLM-generated code? A rigorous & high-standard benchmark (ICLR 2025 · similarity 0.89)
- Training Language Models on Synthetic Edit Sequences Improves Code Synthesis (ICLR 2025 · similarity 0.89)
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.