New WizardLM model, now in 13B! Trained on 250k 'evolved instructions' from ShareGPT and recorded as matching or beating GPT4 on multiple benchmarks (not all, of course :) )