1 paper
Hari Prasad, Ritam Pal
Modern LLM deployments often combine quantization with higher sampling temperatures to reduce cost, latency, or repetition, yet safety evaluations usually treat these as fixed impl…