MiniCPM
An open-source local LLM family for assistants, coding and tool use, with GGUF and MLX downloads under Apache 2.0.
MiniCPM is a family of compact language models for developers building local assistants and coding agents. It targets on-device workloads and hardware with limited computing resources.
- Coding and reasoning: The models support code, mathematics, tool use and agent tasks, with long-context support for larger inputs.
- Model choice: MiniCPM5-1B has a smaller footprint, while MiniCPM5-2B trades a larger footprint for stronger capabilities.
- Local model formats: Downloads include BF16, GGUF and MLX variants, giving developers a choice of formats for local deployment.
- Adaptation: Base and fine-tuned checkpoints are available, alongside UltraData training datasets covering web text, code, agent tasks and reasoning.
MiniCPM5-2B uses the standard LlamaForCausalLM architecture. FlagOS adaptations support multiple chip vendors, including NVIDIA, Ascend and ARM-v9.
The project is open source under Apache 2.0. An online demo is available separately from the downloadable models.
