Tag
An update on the LoRA over GGUF series showing that Qwen3.8-Flash-Next (125B-A6B + 51B engram) can now be LoRA-trained in 40 GiB VRAM without CPU offloading, achieving 200 token/s on Strix Halo, with initial GGUF support merged in transformers 5.18 and further kernel optimizations (GGTensile) on the horizon.
A Windows desktop app called SEELS that allows users to run local LLMs, correct model responses, and automatically train LoRA adapters from corrections. It bundles voice mode, hardware dashboard, and a training sidecar.