Feedback about Failsafe mode:

[cgessai]

Working fine on an old laptop under Win 7 though (predictably) very slow. It's taking about 8 seconds per token but otherwise appears to be working well with WizardLM 7B Uncensored Q4_0 in both Instruct and Story mode. FWIW, the model has been quantized using the newer May 19th format. Also, using command line args: --noavx2 --noblas --nommap --stream --launch and url with ?streamamount=2

[LostRuins]

Great glad to know it works! This will allow even those with ancient devices to try it out.