Athene-70B
Nexusflow · 2024-07-19 · 70.6B parameters
Athene-70B is an open-weight chat model produced by Nexusflow by applying reinforcement learning from human feedback to Llama 3 70B Instruct. It was released as a general-purpose conversational model with the Llama 3 chat template and an 8,192-token context window.
Benchmark scores
| Benchmark | Score |
|---|---|
| BigCodeBench Complete | 47.5 |
| BigCodeBench Instruct | 36.8 |
| BigCodeBench-Hard Complete | 24.3 |
| BigCodeBench-Hard Instruct | 15.5 |
| ZeroEval | 41.2 |
| ZeroEval CRUX | 50.6 |
| ZeroEval GSM8K | 86.7 |
| ZeroEval MATH Level 5 | 20.7 |
| ZeroEval MMLU-Redux | 76.6 |
| ZeroEval ZebraLogic | 16.7 |