InternLM
An open-source LLM family with reasoning and chat models, Transformers support, and a long-context variant that accepts up to one million tokens.
InternLM is a family of downloadable language models for developers and researchers building their own AI applications. It includes models for general conversation, complex reasoning and coding, with separate base and chat variants for customization or use in an assistant.
InternLM3-8B-Instruct supports ordinary conversational responses and a deep thinking mode that uses longer reasoning chains for difficult problems. InternLM2.5-Chat focuses on following instructions, conversation and function calling. Its Chat-1M variant accepts up to one million tokens of context, useful when an application needs to work with lengthy documents or large amounts of reference material. The family also supports internet search and combining information gathered across web pages; those tasks need internet access.
The models work with Hugging Face Transformers, with Python and PyTorch as part of the software stack. LMDeploy provides deployment and model serving within the surrounding toolchain. Developers can choose smaller models for lighter applications or larger variants for more complex tasks, rather than committing to one model size.
The repository uses the Apache 2.0 license. For teams adapting models, the ecosystem includes XTuner for training and fine-tuning, plus OpenCompass for evaluation. InternLM2-Reward provides separate models for assessing response preferences in reinforcement learning workflows.
