Skip to content
Chapter 7 · Decisions 7-5

Retire the legacy Llama line

Ru's conversational generation is cloud-first. Llama 3.1/v6.x and the later LFM2.5 local-generation/training line are removed after a path/reference/ protection manifest because retaining unsupported local assets obscures the active runtime and produces misleading operations documentation.

The tracked local runtime, training tooling, evaluator, configs, and generated local assets were removed on 2026-08-21. A narrow settings migration remains so old persisted llama.cpp/GGUF values fall back to the cloud default; it does not load or retain a local model.