You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
As opposed to the merge_lora.py which allows for merging into a checkpoint, svd_merge_lora.py is strictly for standalone merges. But, it supports both mismatched dims, and LyCORIS merging. Two things sorely lacking from merge_lora.py, and from this GUI.
To make things easier on you from a dev standpoint, it supports the same --precision, --save_precision, --models, --ratios, --save_to args as the normal merge_lora.py. It drops --sd_model for obvious reasons, as well as --v2, presumably also because it's not loading a checkpoint. It adds --new_rank and --new_conv_rank for LoRA rank and Conv2D rank respectively, which can work like your rank sliders in other tabs (e.g. fixed locon extraction), and --device, which can also work like it does in other tabs.
Thanks!
The text was updated successfully, but these errors were encountered:
As opposed to the merge_lora.py which allows for merging into a checkpoint, svd_merge_lora.py is strictly for standalone merges. But, it supports both mismatched dims, and LyCORIS merging. Two things sorely lacking from merge_lora.py, and from this GUI.
To make things easier on you from a dev standpoint, it supports the same --precision, --save_precision, --models, --ratios, --save_to args as the normal merge_lora.py. It drops --sd_model for obvious reasons, as well as --v2, presumably also because it's not loading a checkpoint. It adds --new_rank and --new_conv_rank for LoRA rank and Conv2D rank respectively, which can work like your rank sliders in other tabs (e.g. fixed locon extraction), and --device, which can also work like it does in other tabs.
Thanks!
The text was updated successfully, but these errors were encountered: