how change "with on_the_fly = False" to true.

Nobody has claimed this yet.

Hi, I usually work with an RTX 4070, and a few weeks ago I bought a laptop with an RTX 5070. But I've noticed that the images don't come out the same. I checked the settings, uninstalled and reinstalled, but everything seems fine. The only thing I noticed is that when I create an image, a different message appears on both computers.

on the RTx 4070:
[LORA] Loaded H:\stable\Flux\webui\models\Lora\characters\Lady_Nagant-My_Hero_Academia.safetensors for KModel-CLIP with 264 keys at weight 0.8 (skipped 0 keys) **with on_the_fly = True**
and on the RTx 5070:
[LORA] Loaded H:\stable\Flux\webui\models\Lora\characters\Lady_Nagant-My_Hero_Academia.safetensors for KModel-CLIP with 264 keys at weight 0.8 (skipped 0 keys) **with on_the_fly = False**

so I think the inconsistency has to do with that parameter **bnb-nf4** Both computers create the same image

Start by reproducing the LoRA load on the RTX 4070 and RTX 5070 with the same settings, then compare the runtime configuration around bnb-nf4 and the on_the_fly value. Done means identifying why the log values differ and verifying whether bnb-nf4 provides a supported way to control that behavior.

Written by the indexing model from the issue text.