NEW WizardLM-2 8x22B: Fine-tune & Stage-DPO align

New LLM from Microsoft: WizardLM-2 8x22B, based on Mixtral 8x22B (by Mistral AI), but further fine-tuned and stage-DPO aligned plus RLEIF (instruction quality reward model (IRM) combined with the process supervision reward model (PRM)). Open-source?

HuggingFace link to WizardLM-2 8x22B LLM (fine-tuned and aligned):
https://huggingface.co/microsoft/WizardLM-2-8x22B

#airesearch
#airesearch
#microsoft

huggingface llm microsoft mistral mistral ai mixtral mixtral 8x22b process process supervision quality reward model stage supervision