Review Generation on FIRSTPASS n = 372 (test)
2,155Review LengthHuman reference
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Human reference2026.06 | 2,155 | — | — | — | — | |
| FIRSTPASSProtocol=SFT2026.06 | 1,187 | 35.4 | 8.8 | 15.4 | — | |
| Qwen2.5-7BProtocol=zero-shot2026.06 | 1,019 | 37.3 | 9 | 13.6 | 0.001 | |
| Llama-3-8BProtocol=zero-shot2026.06 | 1,006 | 39.1 | 10.2 | 16.4 | 0.001 | |
| Gemini-3.1-flash-lite-previewProtocol=ZS2026.06 | 934 | 39.9 | 10.9 | 14.6 | 0.001 | |
| DeepSeek-R1-7BProtocol=zero-shot2026.06 | 538 | 27.7 | 6.8 | 11.6 | 0.001 |