BebeLM
A pure-Rust, CPU-only library and command-line tool for running LiquidAI's LFM2.5-8B-A1B language model locally, without a GPU.
BebeLM is a Rust library crate, published on crates.io, that implements CPU-only inference for the LFM2.5-8B-A1B Q4_K_M model. The model activates only about 1 billion parameters at a time, so it can generate text at interactive speed without a GPU, needing roughly 6 to 8 GB of RAM depending on context length. The library has intentionally few dependencies and needs no extra system packages to build.
It can be imported into other Rust projects or used through a bundled command-line interface with one-shot completion, interactive multi-turn chat with persistent key-value and conversation caches, and single-turn question answering, all of which stream tokens as they are produced. It was tested on an M5 CPU as well as Ryzen and Threadripper processors, and is expected to run on Intel CPUs and Raspberry Pi 4 or 5, though that is untested.