My first LORA, Calliope Mori from Hololive Myth. Keywords are "mori calliope" and "mori calliope nsfw" (you know the difference).
#Calliope Mori Lora
zenith dune
bleak sleet
Will try as soon as
craggy kelp
bleak sleet
My first LORA, Calliope Mori from Hololive Myth. Keywords are "mori calliope" an...
Did you delete it?
steep matrix
Did you delete it?
probably civitai instability during chinese hours.
late basin
Civitai's server weakness is too bad...
How about uploading a backup to Huggingface?
glossy moss
Now to make some lewds
zenith dune
Will get onto it soon, about to head home to pc. (It was set to draft, changed to published now)
bleak sleet
CLYDE PLEASE
IT'S JUST A BIKINI
anyway pretty good lora, better than the one floating around outside civitai
CLYDEEEE
@zenith dune I compressed your lora from 144MB to 37MB, that's the result, would you like to have it
clyde?
zenith dune
Thanks! Yes, would be interested. I've been slowly looking into file sizes of loras, couldn't understand why they're coming out so large compared to a TI
bleak sleet
bleak sleet
Thanks! Yes, would be interested. I've been slowly looking into file sizes of lo...
The default when training is dim128 or dim64, I forgot, while this one is dim16
Even dim8 and fp16 (so like, 8 MB) is good enough for most loras or so I heard, but most people don't know that
I didn't want to compress it further than that, it's best if they're trained like that in the first place
zenith dune
How did you compress it after tuning? Have been looking into dim sizes default on the easy training scripts was 128 i believe
bleak sleet
On this
https://github.com/derrian-distro/LoRA_Easy_Training_Scripts
You use lora_resize.bat and it lets you choose what you want
zenith dune
Ah thank you
bleak sleet
If you know some programming this might make sense to you, I edited the base script from there to convert a whole folder at the same time
import argparse
import os
from tkinter import filedialog
from tkinter import simpledialog
from tkinter import messagebox
import networks.resize_lora as resize
INPUT = "./resize_input/"
OUTPUT = "./resize_output/"
def main():
print("Rezising all in "+INPUT)
for filename in os.listdir(INPUT):
print(f"\nResizing {filename}\n")
resize_file(16, INPUT+filename, OUTPUT+filename)
def resize_file(dim: int, input: str, output: str):
parser = argparse.ArgumentParser()
parser.add_argument("--save_precision", type=str, default=None, choices=[None, "float", "fp16", "bf16"])
parser.add_argument("--new_rank", type=int, default=4)
parser.add_argument("--save_to", type=str, default=None)
parser.add_argument("--model", type=str, default=None)
parser.add_argument("--device", type=str, default=None)
parser.add_argument("--verbose", action="store_true")
args_list = ["--device=cuda", f"--new_rank={dim}", f"--model={input}", f"--save_to={output}"]
args = parser.parse_args(args_list)
resize.args = args
resize.resize(args)
if __name__ == "__main__":
main()
I know enough to compare them side-by-side
bleak sleet
fp16 also doesnt change much
Could you check for me if the default is fp16 to begin with?
zenith dune
Seems it is fp16 by default, yeah
harsh gulch
worst music ever
