Fluency on BOOKS
4.46Fluency ScorePROXIMALByte
Evaluation Results
| Method | Links | |
|---|---|---|
| PROXIMALByteSafe model (ps)=COMMA 7B, Risky model (pr)=LLAMA 4 17B×16E, Decoding level=Byte-level, NCR threshold=>= 75%2026.02 | 4.46 | |
| PROXIMALByteSafe model (ps)=TINYCOMMA 1.8B, Risky model (pr)=LLAMA 4 17B×16E, Decoding level=Byte-level, NCR threshold=>= 75%2026.02 | 4.43 | |
| CP-FUSESafe model (ps)=COMMA 7B, Risky model (pr)=LLAMA 4 17B×16E, Decoding level=Byte-level, NCR threshold=>= 75%2026.02 | 4.3 | |
| PROXIMALByteSafe model (ps)=COMMA 7B, Risky model (pr)=QWEN 2.5 72B, Decoding level=Byte-level, NCR threshold=>= 75%2026.02 | 4.3 | |
| PROXIMALByteSafe model (ps)=TINYCOMMA 1.8B, Risky model (pr)=QWEN 2.5 72B, Decoding level=Byte-level, NCR threshold=>= 75%2026.02 | 4.27 | |
| PROXIMALByteSafe model (ps)=COMMA 7B, Risky model (pr)=LLAMA 3.1 70B, Decoding level=Byte-level, NCR threshold=>= 75%2026.02 | 4.23 | |
| SAFESafe model (ps)=COMMA 7B, Risky model (pr)=LLAMA 3.1 70B, Decoding level=Byte-level, NCR threshold=>= 75%2026.02 | 4.1 | |
| SAFESafe model (ps)=COMMA 7B, Risky model (pr)=QWEN 2.5 72B, Decoding level=Byte-level, NCR threshold=>= 75%2026.02 | 4.1 | |
| SAFESafe model (ps)=COMMA 7B, Risky model (pr)=LLAMA 4 17B×16E, Decoding level=Byte-level, NCR threshold=>= 75%2026.02 | 4.1 | |
| PROXIMALSafe model (ps)=TINYCOMMA 1.8B, Risky model (pr)=LLAMA 3.1 70B, Decoding level=Token-level, NCR threshold=>= 75%2026.02 | 4.02 | |
| TOKENSWAPSafe model (ps)=COMMA 7B, Risky model (pr)=QWEN 2.5 72B, Decoding level=Byte-level, NCR threshold=>= 75%2026.02 | 3.98 | |
| CP-FUSESafe model (ps)=TINYCOMMA 1.8B, Risky model (pr)=LLAMA 4 17B×16E, Decoding level=Byte-level, NCR threshold=>= 75%2026.02 | 3.97 | |
| CP-FUSESafe model (ps)=COMMA 7B, Risky model (pr)=QWEN 2.5 72B, Decoding level=Byte-level, NCR threshold=>= 75%2026.02 | 3.95 | |
| TOKENSWAPSafe model (ps)=COMMA 7B, Risky model (pr)=LLAMA 3.1 70B, Decoding level=Byte-level, NCR threshold=>= 75%2026.02 | 3.9 | |
| TOKENSWAPSafe model (ps)=TINYCOMMA 1.8B, Risky model (pr)=QWEN 2.5 72B, Decoding level=Byte-level, NCR threshold=>= 75%2026.02 | 3.8 | |
| TOKENSWAPSafe model (ps)=TINYCOMMA 1.8B, Risky model (pr)=LLAMA 3.1 70B, Decoding level=Token-level, NCR threshold=>= 75%2026.02 | 3.77 | |
| CP-FUSESafe model (ps)=COMMA 7B, Risky model (pr)=LLAMA 3.1 70B, Decoding level=Byte-level, NCR threshold=>= 75%2026.02 | 3.75 | |
| TOKENSWAPSafe model (ps)=TINYCOMMA 1.8B, Risky model (pr)=LLAMA 4 17B×16E, Decoding level=Byte-level, NCR threshold=>= 75%2026.02 | 3.75 | |
| RCADSafe model (ps)=TINYCOMMA 1.8B, Risky model (pr)=LLAMA 4 17B×16E, Decoding level=Byte-level, NCR threshold=>= 75%2026.02 | 3.48 | |
| RCADSafe model (ps)=COMMA 7B, Risky model (pr)=LLAMA 4 17B×16E, Decoding level=Byte-level, NCR threshold=>= 75%2026.02 | 3.48 | |
| RCADSafe model (ps)=COMMA 7B, Risky model (pr)=LLAMA 3.1 70B, Decoding level=Byte-level, NCR threshold=>= 75%2026.02 | 3.46 | |
| RCADSafe model (ps)=TINYCOMMA 1.8B, Risky model (pr)=LLAMA 3.1 70B, Decoding level=Token-level, NCR threshold=>= 75%2026.02 | 3.38 | |
| CP-FUSESafe model (ps)=TINYCOMMA 1.8B, Risky model (pr)=QWEN 2.5 72B, Decoding level=Byte-level, NCR threshold=>= 75%2026.02 | 3.26 | |
| CP-FUSESafe model (ps)=TINYCOMMA 1.8B, Risky model (pr)=LLAMA 3.1 70B, Decoding level=Token-level, NCR threshold=>= 75%2026.02 | 3.21 | |
| MEMFREESafe model (ps)=TINYCOMMA 1.8B, Risky model (pr)=LLAMA 3.1 70B, Decoding level=Token-level, NCR threshold=>= 75%2026.02 | 3.18 | |
| SAFESafe model (ps)=TINYCOMMA 1.8B, Risky model (pr)=LLAMA 3.1 70B, Decoding level=Token-level, NCR threshold=>= 75%2026.02 | 3 | |
| SAFESafe model (ps)=TINYCOMMA 1.8B, Risky model (pr)=QWEN 2.5 72B, Decoding level=Byte-level, NCR threshold=>= 75%2026.02 | 3 | |
| SAFESafe model (ps)=TINYCOMMA 1.8B, Risky model (pr)=LLAMA 4 17B×16E, Decoding level=Byte-level, NCR threshold=>= 75%2026.02 | 3 |