Nous Hermes 2 Yi 34B
Nous Research · 2023-12-25 · 34.4B parameters
Nous Hermes 2 Yi 34B is Nous Research's instruction-tuned fine-tune of the Yi-34B base model, trained on 1,000,000 entries of primarily GPT-4-generated data plus other high-quality data from open datasets. It uses ChatML for multi-turn dialogue and, at release, Nous reported that it outperformed all earlier Nous Hermes and Open Hermes models across its benchmark comparisons.
Benchmark scores
| Benchmark | Score |
|---|---|
| EQ-Bench v2 | 72.7 |
| EQ-Bench v2 + MAGI-Hard Combined | 67.9 |
| MAGI-Hard | 63.0 |