WizardLM 2 8x22B

The WizardLM-2 7B model, developed by WizardLM@Microsoft AI, is a fast and efficient variant of Microsoft AI's latest Wizard model. By finetuning Mistral 7B Instruct, it achieves comparable performance to models 10 times its size. This multilingual model combines speed with high reasoning capabilities, making it a top choice for various tasks. Utilizing a synthetic training system, WizardLM-2 models have undergone rigorous evaluations. They have shown competitive performance in MT-Bench evaluations, surpassing leading baselines at different model scales. Human Preferences Evaluations have also highlighted the strengths of WizardLM-2 models in various tasks such as writing, coding, math, and reasoning. The WizardLM-2 7B supports multi-turn conversations and adopts the prompt format from Vicuna.

Model card on Hugging Face