OneBench
Greedy Decoding Is Not Precision-Invariant: Cross-Precision Output Divergence in LLM Inference | OneBench: AI Insights