Quantized code LLMs appear more robust than full-precision ones in a majority of tested adversarial and noise scenarios, but the proposed Relative Robustness Score is misspecified.
Deepseek coder 1.3b base bnb 4bit,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.SE 1years
2025 1verdicts
REJECT 1roles
background 1polarities
support 1representative citing papers
citing papers explorer
-
Smaller = Weaker? Benchmarking Robustness of Quantized LLMs in Code Generation
Quantized code LLMs appear more robust than full-precision ones in a majority of tested adversarial and noise scenarios, but the proposed Relative Robustness Score is misspecified.