Full fine-tuning yields statistically different and more focused attribution patterns than LoRA or quantized LoRA, while larger models prioritize numerical constraints and rule identifiers but show performance plateaus beyond 7B parameters in semantic similarity for code compliance.
Median F1 and F3 CodeBERTScore demonstrate an upward trend from the 0.5B model to the ~7B models
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
LLM attribution analysis across different fine-tuning strategies and model scales for automated code compliance
Full fine-tuning yields statistically different and more focused attribution patterns than LoRA or quantized LoRA, while larger models prioritize numerical constraints and rule identifiers but show performance plateaus beyond 7B parameters in semantic similarity for code compliance.