Standard speculative sampling acceptance length and speedup on Vicuna-13B (temperature=0)
2.18 tokens_per_cycle
Fields
id
benchmarks/sd-sps-acceptance-vicuna13b-t0Metric
quality
Value
2.18
Unit
tokens_per_cycle
Model
Vicuna-13B
Methodology
temperature=0, 6 tasks. Average acceptance length τ = 2.18 tokens per draft-verify cycle. Speedup 1.88x mean across tasks.
Measured by
self
Reproducible
yes
notes
Standard speculative sampling with Vicuna-68M as draft model. Chain-structured draft. Lossless (distribution-preserving). Baseline is vanilla autoregressive decoding.
Notes
Standard speculative sampling with Vicuna-68M as draft model. Chain-structured draft. Lossless (distribution-preserving). Baseline is vanilla autoregressive decoding.