Empirical study on five LLMs finds pretrained-to-aligned paths yield bigger gains over baseline than finetuned-to-aligned paths, though absolute accuracy remains lower for pretrained starts.
Title resolution pending
2 Pith papers cite this work, alongside 2 external citations. Polarity classification is still indexing.
2
Pith papers citing it
2
external citations · external index
citation-role summary
background 1
citation-polarity summary
fields
cs.SE 2years
2026 2verdicts
UNVERDICTED 2roles
background 1polarities
background 1representative citing papers
Data-influence-score filtering using validation-set loss on downstream coding tasks improves Code-LLM performance, with the most beneficial training data varying significantly across different programming tasks.
citing papers explorer
-
Reward-Free Code Alignment from Pretrained or Fine-Tuned LLM: Unpacking the Trade-offs for Code Generation
Empirical study on five LLMs finds pretrained-to-aligned paths yield bigger gains over baseline than finetuned-to-aligned paths, though absolute accuracy remains lower for pretrained starts.
-
An Empirical Study on Influence-Based Pretraining Data Selection for Code Large Language Models
Data-influence-score filtering using validation-set loss on downstream coding tasks improves Code-LLM performance, with the most beneficial training data varying significantly across different programming tasks.