Atlas

Models

← All models

Nous Hermes 2 Mixtral 8x7B SFT

Nous Research · 2023-12-26 · 46.7B parameters

Nous Hermes 2 Mixtral 8x7B SFT is Nous Research's supervised-fine-tune-only variant of the Mixtral 8x7B sparse mixture-of-experts model, trained on over one million entries of primarily GPT-4-generated data and other high-quality open datasets. It uses the ChatML prompt format and was released alongside a companion SFT+DPO variant.

Benchmark scores

BenchmarkScore
BBH (Open LLM Leaderboard v2)50.6
Capability110.4
EQ-Bench v272.9
EQ-Bench v2 + MAGI-Hard Combined62.4
GPQA (Open LLM Leaderboard v2)30.2
IFEval (Open LLM Leaderboard v2)57.3
MAGI-Hard51.8
MATH Level 5 (Open LLM Leaderboard v2)2.1
MMLU-Pro (Open LLM Leaderboard v2)30.7
MuSR (Open LLM Leaderboard v2)42.1
Loading Atlas data…