realoperator42/trinity-nano-stheno-lora (attention-only LoRA, r=16, alpha=32) merged into arcee-ai/Trinity-Nano-Preview and converted to GGUF.

Architecture is afmoe, so you need a llama.cpp build that includes AFMOE support.

The embedded chat template is the adapter's corrected ChatML, with the training-only {% generation %} markers removed so --jinja parses cleanly.