>Zephyr alpha is a Mistral fine-tune that achieves results similar to Chat Llama 70B in multiple benchmarks and above results in MT bench (image below). The average perf across ARC, HellaSwag, MMLU and TruthfulQA is 66.08, compared to Chat Llama 70B's 66.8, Mistral Open Orca 66.08, Chat Llama 13B 56.9, and Mistral 7B 60.45. This makes Zephyr a very good model for its size.
>Zephyr alpha is a Mistral fine-tune that achieves results similar to Chat Llama 70B in multiple benchmarks and above results in MT bench (image below). The average perf across ARC, HellaSwag, MMLU and TruthfulQA is 66.08, compared to Chat Llama 70B's 66.8, Mistral Open Orca 66.08, Chat Llama 13B 56.9, and Mistral 7B 60.45. This makes Zephyr a very good model for its size.
source: https://www.reddit.com/r/LocalLLaMA/comments/174t0n0/hugging...