List your company

Multi-path Speculative Decoding on Llama (test)

14.77Throughput (tokens/s)Traversal
12.357212.983613.6114.2364Feb 19, 2026

Evaluation results

MethodLinks
TraversalExpansion Mode=Static2026.0214.77
SpecInfer NDEExpansion Mode=Neural Dynamic Expansion2026.0214.27
SpecTr NDEExpansion Mode=Neural Dynamic Expansion2026.0214.23
NaiveTree NDEExpansion Mode=Neural Dynamic Expansion2026.0214.01
Khisti NDEExpansion Mode=Neural Dynamic Expansion2026.0213.24
NSS NDEExpansion Mode=Neural Dynamic Expansion2026.0212.45