List your company

Multi-hop Question Answering on 2WikiMultiHopQA (Speedup)

3.28Speedup FactorHybrid Verified Decoding
0.77361.42432.0752.7257May 31, 2026

Evaluation results

MethodLinks
Hybrid Verified DecodingTarget Model=Qwen3-8B, Reference Baseline=SuffixDecoding2026.053.28
Hybrid Verified DecodingTarget Model=Qwen3-8B, Reference Baseline=Greedy Decoding2026.052.73
Hybrid Verified DecodingTarget Model=Llama3.1-8B, Reference Baseline=Greedy Decoding2026.052.52
Hybrid Verified DecodingTarget Model=Llama3.1-8B, Reference Baseline=SuffixDecoding2026.052.49
Hybrid Verified DecodingTarget Model=Qwen3-4B, Reference Baseline=Greedy Decoding2026.052.01
Hybrid Verified DecodingTarget Model=Qwen3-4B, Reference Baseline=SuffixDecoding2026.051.65
Hybrid Verified DecodingTarget Model=Qwen3-8B, Reference Baseline=Frequency-Based Hybrid Decoding2026.051.4
Hybrid Verified DecodingTarget Model=Qwen3-4B, Reference Baseline=EAGLE32026.051.19
Hybrid Verified DecodingTarget Model=Llama3.1-8B, Reference Baseline=EAGLE32026.051.04
Hybrid Verified DecodingTarget Model=Qwen3-4B, Reference Baseline=Frequency-Based Hybrid Decoding2026.050.98
Hybrid Verified DecodingTarget Model=Qwen3-8B, Reference Baseline=EAGLE32026.050.96
Hybrid Verified DecodingTarget Model=Llama3.1-8B, Reference Baseline=Frequency-Based Hybrid Decoding2026.050.87