SpecBench
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
SpecBench
900.7AVG SR
47
SPECBENCH Average over scenarios
93.73Safety Score
33
SpecBench v1 (test)
4.9WRIT τ
12
SpecBench Qwen-2-7B-Instruct (test)
3.65Overall Mean Score
5
SpecBench Task 1
74Improvement Rate (T0->T5)
4
SpecBench Curated Domains Cold-start spec-mode
1CSC
4
SPECBENCH
96.8Safety Score
4
SpecBench Vicuna-33B v1.3 (test)
3.52MT Score
4
SpecBench
116.95Tokens/s
3