Vicente Bertolotti

Context-Dependent Effects of Dynamic Confidence Thresholds for Speculative Decoding on a Consumer GPU

This technical note reports a consumer-GPU case study of generated-string-aware confidence thresholds for DSpark speculative decoding. On a six-source long-context holdout using Qwen3-8B Q4_K_M and an AMD Radeon RX 6800, the dynamic policy reached 1.0224x fixed-DSpark decode thro...

Context-Dependent Effects of Dynamic Confidence Thresholds for Speculative Decoding on a Consumer GPU

This technical note reports a consumer-GPU case study of generated-string-aware confidence thresholds for DSpark speculative decoding. On a six-source long-context holdout using Qwen3-8B Q4_K_M and an AMD Radeon RX 6800, the dynamic policy reached 1.0224x fixed-DSpark decode thro...