rlx-bonsai 0.2.4

Bonsai small-reasoning family runner.

Bonsai (1.7B / 2B / 4B / 8B) ships as general.architecture = llama in its GGUF converters — it's a standard Llama-shaped decoder with tuned hyperparameters for small-context reasoning. This crate is a thin wrapper over [rlx_llama32::Llama32Runner] that:

• validates the GGUF actually has arch = llama (catches users pointing at the wrong file);
• gives rlx-run bonsai its own subcommand for discoverability;
• keeps the door open for Bonsai-specific tuning later (e.g. per-family compile profile, eval prompts) without changing the user-facing API.