InternLM

Why we included this project

The InternLM repo is the official source for the InternLM model line, covering the original 7B and 20B checkpoints through the InternLM2, InternLM2.5, and InternLM3 generations. You'll find base and instruction-tuned chat variants for each release, plus download links and model cards. The most recent InternLM3-8B-Instruct is worth a look if compute is tight: it reaches reasoning and knowledge scores comparable to Llama3.1-8B and Qwen2.5-7B while training on only about 4 trillion tokens, roughly a quarter of what similarly sized peers typically need. The repo also hosts reward models and long-context chat variants that handle up to a million tokens, which helps if you're building retrieval-heavy or long-document applications. For teams working in Chinese-English bilingual settings, this is a well-documented place to start.