Microsoft AI has announced the release of WizardLM-2, an opensource large language model with improved performance in chat, reasoning, and multilingual tasks. The new family includes WizardLM-2 8x22B, 70B, and 7B, with impressive capabilities. Through innovative training methods like Evol-Instruct and AAA, WizardLM-2 achieves top-tier performance among both proprietary and opensource models. Human preference evaluations reveal WizardLM-2’s competitiveness against cutting-edge models like GPT-4. The models are available for use on Huggingface, with detailed prompt guidelines for ensuring high-quality generation. Overall, WizardLM-2 represents a significant advancement in AI language models, setting a new standard for performance and efficiency.
https://wizardlm.github.io/WizardLM2/