2026-11-15 · America/Los_Angeles · 社区动态 · #4
Why download abliterated Flash Next models when LoRA exists?
TL;DR -- Download LoRA, add --lora HUIHUI-V2-Qwen3.8-Flash-Next-LoRA-F32.gguf to your llama.cpp launch command for ANY FN quant, and boom, you've got an identical huihui v2 abliterated model on demand and you saved 150GB of space. I just wanted to post a kind of PSA for everyone out there downloading hundreds of GB of abliterated models. You don't need to do that -- the vast majority of abliteration techniques are rank-1 ablations, meaning their refusal projection can be encapsulated in a tiny LoRA and applied at runtime to a quant you already have on disk. It can even be applied…
热度 47.5 / 100;排名与评分保留该期记录。