all / g / s / fur / art / ai / 3dcg / f / rp / req / p2p / dis / lit / kaki chat
[Settings] [Home] [Catalog] [Search] [Thread list] [Stats] [Reports] [Watcher] [PMs] [Admin]
[Return] [Bottom]

Posting mode: Reply
BBCode
(for deletion)
  • Allowed file types are: gif, jpg, jpeg, png, webp, webm, mp4
  • Maximum file size allowed is 25000 KB.
  • Images greater than 255 * 255 pixels will be thumbnailed.
  • 145 unique users in the last 30 minutes (including lurkers)
    First
    [1]
    		Last
    File: ComfyUI_84753_.png
    (1.35 MB, 1344x768) ▶
    1415957
    No.8986 [Report] AS
    ▶
    Anyone here has a 3090 or a 4090 and is willing to perform a finetune of a guro checkpoint?

    we can learn how to configure the system together.
    >>
    No.8992 [Report]
    ▶
    would a 4080 work?
    >>
    No.9005 [Report]
    ▶
    how much Vram does your 4080 have?
    >>
    No.9008 [Report]
    ▶
    16gb>>9005
    >>
    No.9074 [Report]
    ▶
    probably too little. and it will take several days of continous use. THanks for the offer mate
    >>
    No.9529 [Report]
    ▶
    File: Untitled.png
    (14 KB, 731x244) ▶
    15098
    You're going to have to walk me through it. I'm not very computer literate. By the way, where can I find guro loras for illustrious? I saw a thread >>4693 for uterus removal, and one >>2894 for bisection, but they're not for illustrious/noobai. I'm particularly interested in bisection and beheadings.
    >>
    No.9530 [Report]
    ▶
    Okay, so I've found >>3818. But about the checkpoint, how many images is your dataset? I feel like a checkpoint wouldn't really be worth it over loras if we don't have at least 2k. But idk, I don't know this technology and I've never done this before. What base model are you training from?
    >>
    No.9787 [Report]
    ▶
    Okay, I've got taggui and kohya_ss installed, what now?
    >>
    No.9793 [Report]
    ▶
    >>9529
    >>9787
    Awesome. I'm back and we finaly can make this work.


    I think we need a better way to comunciate. Do you have sessoin or discord?


    You did th ehard part, once kohya_ss is inlace i send yo uthe dataset and you run it
    >>
    No.9850 [Report]
    ▶
    >>9787
    >>9529
    I feel so excited to be able to complete this porject now. please do not disappear anon.
    >>
    No.9852 [Report]
    ▶
    Right now I've only a 5500, and the second 3090 is running at like x4, which is fine for inference... I've ordered a used 3945wx, which should let me use both 3090s at full speed (for training, I want x16 on both cards right or they'll bottleneck right?). But it'll take over a month to arrive, rippy.
    >>
    No.9853 [Report]
    ▶
    >>9794
    I've send a request on discord, but I'll be busy tomorrow. Might take me some time to respond.
    >>
    No.9874 [Report]
    ▶
    >>9852
    make sure your mobo supports dual x16 too
    >>
    No.9875 [Report]
    ▶
    >>9794
    i have also sent a discord request (i was the guy with the 3080). if you are planning on training an ayaswan dataset, i have a pretty complete collection of his uncen works and i want to see if i can contribute anything
    >>
    No.9876 [Report]
    ▶
    >>9874
    I'm pretty sure all wrx80 boards have like 7 x16s lol
    >>9875
    3080? Not a 4080?
    >>
    No.9877 [Report]
    ▶
    >>9875
    Have they already been tagged?
    >>
    No.9878 [Report]
    ▶
    86 days for 10 epochs... I think something's fucky
    >>
    No.9880 [Report]
    ▶
    I guess a full finetune of a model is too much for amateurs. Going to try a lora instead.
    >>
    No.9889 [Report]
    ▶
    >>9877
    yes. I have 6000 manually tagged guro images. divided by artist.

    If you want write me on discord and i wills hare it with you.
    [Top]
    Delete post: []
    First
    [1]
    		Last