Instruction-Tuning-with-GPT-4/GPT-4-LLM
A dataset release of instruction-following data generated by GPT-4 for fine-tuning language models via supervised and reinforcement learning.
Velocity · 7d
+3.7
★ / day
Trend
→steady
star history
The GPT-4-LLM project releases instruction-following datasets generated by GPT-4 for training LLMs. It provides English and Chinese instruction data created using Alpaca prompts and Chinese translations via ChatGPT. The repository also includes comparison data ranked by GPT-4 for training reward models and answers on Unnatural Instructions for quantifying gaps between GPT-4 and instruction-tuned models.