PicoLM
Open source
FreeFree forever
Overview
PicoLM is a minimal LLM runtime designed for edge devices with very limited memory and processing power. Written in approximately 2500 lines of C11 with zero external dependencies, it loads GGUF models directly from disk using memory mapping and can run TinyLlama 1.1B on devices with as little as 256MB RAM. It supports JSON constrained generation, multiple quantization formats, ARM NEON and x86 SIMD optimizations and integration with local AI agents such as PicoClaw.
Key features
- Pure C implementation
- Runs on 256MB RAM
- GGUF model support
- JSON constrained generation
