scheduled july releases from the WizardLM team:

- WizardLM-2 Paper & Model
- Evol-Instruct V2 Paper
- Wizard Agent Paper
- Arena-Learning Paper

☺︎

Auto Evol-Instruct
paper: https://arxiv.org/pdf/2406.00770

“Auto Evol enables WizardLM2 to be trained using a virtually unlimited number and variety of synthetic data.” 😈

✔️Auto Evol-Instruct enhances instruction data through an automated two-stage process. first, the optimizer LLM identifies and analyzes issues in evolving instructions. then, it refines these methods to produce a diverse and complex dataset, improving instruction quality.

✔️Auto Evol-Instruct broadens WizardLM-2. the fully automated Evol-Instruct pipeline has expanded WizardLM-2 from the three initial domains (chat, code, and math) of WizardLM-1 to include new domains.