Research map / Large language models / Instruction and fine-tuning
TFix: Learning to Fix Coding Errors with a Text-to-Text Transformer
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 |
|---|---|---|---|
| Add natural language pre-training (T5 init vs. from scratch) (ablation) | T5-large fine-tuned on code fixes · clean test | exact match: 49.3 vs 6.7 | Helped |
| Fine-tune all 52 error types jointly instead of per type (ablation) | T5-large · clean test | exact match: 49.3 vs 36.3 | Helped |
| Fine-tune all error types jointly instead of per type (ablation) | T5-large · random test | error removal: 52.5 vs 34.2 | Helped |
| Scale model from T5-small to T5-base (ablation) | T5 fine-tuned on code fixes · clean test | exact match: 39.2 -> 48.5 | Helped |
| Scale model from T5-base to T5-large (ablation) | T5 fine-tuned on code fixes · clean test | exact match: 48.5 -> 49.3 | Helped |
| Use TFix (T5-large) instead of SequenceR/CoCoNuT | code error fixing · restricted test and clean test | exact match: 46.3 vs 23.6 (restricted); 49.3 vs 17.9 (clean) | Helped |
Closest papers by meaning
- Counterexample Guided Program Repair Using Zero-Shot Learning and MaxSAT-based Fault Localization (AAAI 2025 · similarity 0.85)
- LeDex: Training LLMs to Better Self-Debug and Explain Code (NeurIPS 2024 · similarity 0.83)
- LILO: Learning Interpretable Libraries by Compressing and Documenting Code (ICLR 2024 · similarity 0.83)
- Teaching Large Language Models to Self-Debug (ICLR 2024 · similarity 0.82)
- Verifiers in the Loop: Decoding Time Verification for Code Translation (NeurIPS 2026 · similarity 0.82)
- AuPair: Golden Example Pairs for Code Repair (ICML 2025 · similarity 0.82)
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.