Structured Generation on C++
1.42Mean Generation Latency (s)Llama-3B
Evaluation Results
| Method | Links | |
|---|---|---|
| Llama-3BConstraint=Unconstrained2026.05 | 1.42 | |
| Llama-3BConstraint=+CFGZIP2026.05 | 1.47 | |
| Llama-3BConstraint=XGrammar22026.05 | 2.14 | |
| GPT-20BConstraint=Unconstrained2026.05 | 2.34 | |
| GPT-20BConstraint=+CFGZIP2026.05 | 2.41 | |
| Qwen-4BConstraint=Unconstrained2026.05 | 2.48 | |
| Qwen-4BConstraint=+CFGZIP2026.05 | 2.55 | |
| GPT-20BConstraint=XGrammar22026.05 | 4.59 | |
| Qwen-4BConstraint=XGrammar22026.05 | 5.14 |