Spec-Bench
Emerging8papers using it
2025first seen
Spec-Bench is a benchmark dataset used to evaluate the performance of speculative decoding methods in large language models by measuring their inference speed and efficiency.
Papers using Spec-Bench (8)
- RACER: Retrieval-Augmented Contextual Rapid Speculative DecodingWhen, What, and How: Rethinking Retrieval-Enhanced Speculative DecodingS$^4$C: Speculative Sampling with Syntactic and Semantic Coherence for Efficient Inference of Large Language ModelsMirror Speculative Decoding: Breaking the Serial Barrier in LLM
InferenceReasoning over Boundaries: Enhancing Specification Alignment via
Test-time DelibrationBatch Speculative Decoding Done RightToken-Driven GammaTune: Adaptive Calibration for Enhanced Speculative DecodingLossless Acceleration of Large Language Models with Hierarchical
Drafting based on Temporal Locality in Speculative Decoding