List your company

Multi-path Speculative Decoding on Qwen (test)

22.54Throughput (tokens/s)SpecInfer NDE
18.244819.359920.47521.5901Feb 19, 2026

Evaluation results

MethodLinks
SpecInfer NDEExpansion Mode=Neural Dynamic Expansion2026.0222.54
SpecTr NDEExpansion Mode=Neural Dynamic Expansion2026.0221.74
NaiveTree NDEExpansion Mode=Neural Dynamic Expansion2026.0221.67
TraversalExpansion Mode=Static2026.0221.35
NSS NDEExpansion Mode=Neural Dynamic Expansion2026.0220.29
Khisti NDEExpansion Mode=Neural Dynamic Expansion2026.0218.41