The ggml‑org/llama.cpp project released version b11773 on 2026‑10‑21. The update adds a new model type called H2O‑Lightning‑4B, which is a Qwen3.5‑4B model with a merged decision LoRA. The model returns answers prefixed with a space (" A") and can handle up to 255 labels. The converter now writes decision metadata and checks the serve_config.json for a match. The change was co‑authored by H2O AI staff and tested on a range of CPU and GPU platforms.
Why it matters
Developers can now run decision‑type AI models directly with llama.cpp on many hardware setups.
