GGUF quants of a LoRA finetune of the base model, merged in bf16 before quantisation so only one lossy step is applied.

Built with an importance matrix calibrated on the finetuning data (agentic coding traces), hence the i1- prefix.