SpecBench
Benchmarks
| Task Name | Dataset Name | SOTA Result | Trend | |
|---|---|---|---|---|
| Speculative Decoding | SpecBench | AVG SR900.7 | 47 | |
| Specification Alignment | SPECBENCH Average over scenarios | Safety Score93.73 | 33 | |
| Speculative Decoding | SpecBench v1 (test) | WRIT τ4.9 | 12 | |
| Speculative Decoding | SpecBench Qwen-2-7B-Instruct (test) | Overall Mean Score3.65 | 5 | |
| Preference Elicitation | SpecBench Task 1 | Improvement Rate (T0->T5)74 | 4 | |
| Relational Data Synthesis | SpecBench Curated Domains Cold-start spec-mode | CSC1 | 4 | |
| Specification Alignment | SPECBENCH | Safety Score96.8 | 4 | |
| Speculative Decoding | SpecBench Vicuna-33B v1.3 (test) | MT Score3.52 | 4 | |
| LLM Generation | SpecBench | Tokens/s116.95 | 3 |
Showing 9 of 9 rows