Hermes 2 Theta Llama 3 70B
Nous Research · 2024-06-20 · 70.6B parameters
Hermes 2 Theta Llama 3 70B is Nous Research's 70B chat model built by merging Hermes 2 Pro with Meta's Llama-3 Instruct in collaboration with Charles Goddard and Arcee AI, then further refining the result with RLHF. It uses ChatML for multi-turn dialogue and was specifically trained for function calling and JSON structured outputs.
Benchmark scores
| Benchmark | Score |
|---|---|
| BigCodeBench Complete | 55.6 |
| BigCodeBench Instruct | 45.6 |
| BigCodeBench-Hard Complete | 24.3 |
| BigCodeBench-Hard Instruct | 20.3 |
| GPQA Diamond | 37.5 |
| MATH Level 5 | 22.7 |
| OTIS Mock AIME 2024-2025 | 2.5 |