Jericho Siahaya · Open weights, local
Naluri-308M by Jericho Siahaya scores 61.0 on the System One Index (22nd of 24). Calibration error 5.7%.
| Dataset | Accuracy | Rank | Best on the board | Calibration error |
|---|---|---|---|---|
| AG News English | 79.5 | 23rd of 24 | 92.5 | 8.3% |
| SST-2 English | 68.5 | 24th of 24 | 97.0 | 3.8% |
| SST-5 English | 28.0 | 24th of 24 | 62.5 | 32.5% |
| DAIR Emotion English | 84.0 | 3rd of 24 | 91.5 | 15.5% |
| Banking77 English | 27.0 | 21st of 24 | 98.5 | 9.5% |
| BoolQ English | 56.5 | 23rd of 24 | 92.5 | 19.1% |
| XNLI English English | 76.0 | 20th of 24 | 93.3 | 4.7% |
| Typed decisions English | 54.7 | 19th of 24 | 79.0 | 6.5% |
| XNLI French French | 68.0 | 17th of 24 | 90.7 | 11.1% |
| XNLI Arabic Arabic | 56.7 | 22nd of 26 | 89.3 | 14.3% |
| MASSIVE scenario Arabic | 63.5 | 12th of 26 | 89.5 | 11.1% |
| MASSIVE intent Arabic | 65.5 | 19th of 26 | 91.5 | 32.2% |
| OSACT4 offensive Arabic | 79.5 | 22nd of 26 | 95.5 | 5.8% |
| Tweet sentiment Arabic | 58.0 | 17th of 26 | 80.0 | 6.1% |
| MIRACL rerank Arabic | 34.0 | 23rd of 26 | 90.5 | 8.4% |
| MIRACL pairwise Arabic | 76.0 | 21st of 26 | 91.0 | 2.4% |
jerichosiahaya/Naluri-308M-encoderTrained on the train splits of XNLI, MASSIVE and DAIR Emotion; we test their test splits. CC BY-NC 4.0. Run with serve.py from the Naluri repo.
Ranks count the models charted on the leaderboard. Every model gets the same cases and the same request, under the benchmark rules.