Domain-adapted COBOL-Coder LLM achieves 73.95% compilation success and 49.33 Pass@1 on code generation benchmarks, far exceeding GPT-4o and open-source models.
- STOP RUN to return from main (no System.exit unless COBOL implies abnormal termination)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.SE 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
COBOL-Coder: Domain-Adapted Large Language Models for COBOL Code Generation and Translation
Domain-adapted COBOL-Coder LLM achieves 73.95% compilation success and 49.33 Pass@1 on code generation benchmarks, far exceeding GPT-4o and open-source models.