InternLM

Abstract

Pre-training a bilingual 100B Foundation model on data with over a trillion tokens, the model exhibits excellent performance in scenarios such as Chinese, English, and coding due to the appropriate data ratio. Based on the foundation model, the application of high-quality human annotated dialogue data combined with RLHF technology enables the InternLM large language model to respond to complex commands during human interaction, while also demonstrating responses in line with human morality and values.