🏞｜general-with-images

[wispy ether]

or do i need this

[strong sonnet]

See this, if we go to 150, the details are better sorted and fixed, 150 lets you integrate the features better with other elements
https://www.reddit.com/r/StableDiffusion/comments/xb8h58/matrix_comparison_of_cfg_scale_vs_step_count/

[sterile temple]

what version of torch does it show at the bottom of the web ui page?

[wispy ether]

i use 2.0.1

[smoky oak]

I mean yeah, if you are doing irresponsibly high CFG, sure

[wispy ether]

cud 118

[strong sonnet]

I dont use high CFG I either use 7(for faces and realistic images) 11(if I want to get a ficitional image)

[smoky oak]

@sterile templeSorry to ping again, but what would you recommend between these 2? I have no diea what the difference is

[sterile temple]

extract the zip into the extension's directory

[sterile temple]

I picked GA

[strong sonnet]

But it might be a huge diff, but i have some fomo kinda thing so I just max out

[smoky oak]

alright then I guess

[wispy ether]

the one im currently downloading?

[sterile temple]

yes

[smoky oak]

All of my stuff works just fine at 15 steps, so thats what I do. Much faster to itterate on, and much less power wasted

[strong sonnet]

Does anyone know how to automate minigpt4

[strong sonnet]

do you generate realisitic or artisitc images/fictional?

[smoky oak]

cool, downloading

[smoky oak]

both

[strong sonnet]

ok

I have used 20 steps for realistic and many elements were off they looked patchy

[wispy ether]

why

[smoky oak]

15 step realism

[wispy ether]

sd is using 10gb ram

[wispy ether]

and now

[strong sonnet]

they are great

which model do you use?

[wispy ether]

hitler image?

[smoky oak]

A custom mix

[wispy ether]

can you try

[sterile temple]

move that folder into the stable-diffusion-webui-tensorrt folder

[wispy ether]

uh

i dont have

what

[strong sonnet]

@smoky oak I am planning to train a model based of faces and yet another on landscape photography with a huge datset, do you want to collaborate?

[smoky oak]

You are actually insanely annoying

[wispy ether]

oh

[strong sonnet]

or anyone?

[smoky oak]

I would be super down to test and give feedback

I used to hard focus realism, 10's of thousands of gens

[wispy ether]

https://tenor.com/view/angry-annoyed-squidward-spongebob-freak-out-gif-18780512

[wispy ether]

done

and now

[sterile temple]

restart SD

[strong sonnet]

I am planning the faces model with around 500k images and Landscape photography with 100k

[wispy ether]

tf

[strong sonnet]

ok sure, will share in #1047197565365538826 once it's completely out

[smoky oak]

amazing

[strong sonnet]

@smoky oak Do you know how we can automate minigpt

[wispy ether]

tf does tensorrt do to my sd

[smoky oak]

I do not, unfortunately

[wispy ether]

and now

[smoky oak]

you need visual C++, it says right there

[sterile temple]

are you on the dev branch?

[wispy ether]

i have it

someone else needed it before

something*

[wispy ether]

no

what is that

how do i get it

[sterile temple]

close SD

[wispy ether]

done

[sterile temple]

right click in the SD folder, open terminal

[wispy ether]

done

[sterile temple]

git checkout dev

git pull https://github.com/AUTOMATIC1111/stable-diffusion-webui dev

[wispy ether]

@sterile temple

abortion xd

(c) Microsoft Corporation. Alle Rechte vorbehalten.

C:\Users\Lukag\Desktop\stable-diffusion-webui>git checkout dev
error: Your local changes to the following files would be overwritten by checkout:
        launch.py
        webui.py
Please commit your changes or stash them before you switch branches.
Aborting

C:\Users\Lukag\Desktop\stable-diffusion-webui>git pull https://github.com/AUTOMATIC1111/stable-diffusion-webui dev
remote: Enumerating objects: 794, done.
remote: Counting objects: 100% (397/397), done.
remote: Compressing objects: 100% (71/71), done.
remote: Total 794 (delta 335), reused 361 (delta 318), pack-reused 397
Receiving objects: 100% (794/794), 320.65 KiB | 1.21 MiB/s, done.
Resolving deltas: 100% (544/544), completed with 89 local objects.
From https://github.com/AUTOMATIC1111/stable-diffusion-webui
 * branch              dev        -> FETCH_HEAD
error: Your local changes to the following files would be overwritten by merge:
        launch.py
        webui.py
Please commit your changes or stash them before you merge.
Aborting
Updating 89f9faa6..9d953c0e

C:\Users\Lukag\Desktop\stable-diffusion-webui>
C:\Users\Lukag\Desktop\stable-diffusion-webui>```

[smoky oak]

oh damn it

uhhgggg

oh wait

[wispy ether]

wat

[smoky oak]

no nevermind

[sterile temple]

git reset --hard

[wispy ether]

head

[sterile temple]

git pull

[smoky oak]

@sterile templemy SD is downloading pytorch 1.13, any ideas how to get it to do 2.0.1?

Sorry, I know we are both hounding you 😅

[wispy ether]

change the version in the file

also

sd should download 2.0.1 by default

[sterile temple]

i think you add --update-torch to the bat

[wispy ether]

@sterile temple

[sterile temple]

git checkout dev

git pull

[wispy ether]

is it now dev brench?

branch

venv "C:\Users\Lukag\Desktop\stable-diffusion-webui\venv\Scripts\Python.exe"
Python 3.10.11 (tags/v3.10.11:7d4cc5a, Apr 5 2023, 00:38:17) [MSC v.1929 64 bit (AMD64)]
Version: v1.3.1-164-g9d953c0e
Commit hash: 9d953c0e039cce2c9d8318de9b79a09e243d3e45
Fetching updates for K-diffusion...
Checking out commit for K-diffusion with hash: c9fe758757e022f05ca5a53fa8fac28889e4f1cf...
Installing requirements

wat

[smoky oak]

nope, still 1.13 hmmm

well this is a problem lol

[sterile temple]

sorry

[smoky oak]

all good

[sterile temple]

--reinstall-xformers --reinstall-torch

[wispy ether]

im still on 1.2.1 💀

how do i update without destroying sd

like last time

[smoky oak]

thats what I am trying to do too

[wispy ether]

i destroyed it

i mean you can just move the folders of the models etc what you need into the new one

[sterile temple]

did you git pull after git checkout dev?

[wispy ether]

no

[smoky oak]

@dense tapir Are you here, by chance?

[sterile temple]

did you try --reinstall-xformers --reinstall-torch to update?

[sterile temple]

git pull https://github.com/AUTOMATIC1111/stable-diffusion-webui dev

[smoky oak]

the problem is I don't have a venv, so it installs torch 1.13 to try and make it lol

[sterile temple]

ok

[wispy ether]

and now

[sterile temple]

I used these on an existing install and it upgraded okay

[smoky oak]

hmm

alright

[karmic sedge]

I have the feeling that there is still much room for improvement.
RTX 4090 + TensorRT
Without TensorRT I get ~34 it/s

[wispy ether]

i thought without a 4090 gets 24

but wow

i usually get like 6,5it/s on 512x512

https://tenor.com/view/daeth-funi-periflight-dies-from-gif-22212023

[smoky oak]

no lmao

[karmic sedge]

D:

[sterile temple]

I just tried it again, because the first time I tried it was 40% slow.er. But now it's a bit faster

[wispy ether]

@smoky oakdo u still need the tensorrt file

[wispy ether]

what now

[sterile temple]

did you launch SD okay?

[wispy ether]

lets try

hm

well

it booted

but no module named pycuda

[smoky oak]

no

[wispy ether]

ok

[smoky oak]

now I need to figure out how the hell to update my SD to 2.0.1 lmao

[karmic sedge]

768x768, that's quite the improvement.
~15 it/s without TensorRT

[wispy ether]

why not 1.3.1

[smoky oak]

cause --reinstall-torch didn't do anything

[wispy ether]

sd isnt torch

wait

do iu mean torch or the sd version

[wispy ether]

i get 2it/s lol

[smoky oak]

I am probably gonna have to wait for general to get on, as I don't remember how we hard locked my SD to only 1.13.1

[wispy ether]

ok somethiung is wrojg

[karmic sedge]

GTX 1070?

[wispy ether]

3060

ill upgrade maybe soon to a 4070

due to game bottleneck

[sterile temple]

did you hardcode it in the requirements.txt file?

[smoky oak]

oh maybe, let me look

[karmic sedge]

Oh on 768, yea nvm.

[smoky oak]

no

[sterile temple]

I had this. One of the extensions is playing up

[wispy ether]

https://tenor.com/view/kek-zoil-gif-22981488

[smoky oak]

so you're gonna upgrad to a GPU thats like 30% faster

[wispy ether]

ye

and dlss 3

and newer tensor core gen

[smoky oak]

oh right, we already mentioned only reason for 4070 is you are power limited, nevermind

[wispy ether]

yes

and no

i just dont need the best ever pc

[smoky oak]

4070 is kinda bad for AI work, but will still be a lot better than your 3060

[wispy ether]

yes

also because maybe in future fp8

also

4070 is like 200% faster than 3060 in gaming

[sterile temple]

@smoky oak navigate stable-diffusion-webui\venv\Scripts

[karmic sedge]

Which site is that? 😄

[sterile temple]

open terminal

[wispy ether]

like when 3060 gets 45 fps a 4070 gets like 90

[wispy ether]

https://pc-builds.com/bottleneck-calculator/result/1fE1fl/1/general-tasks/3440x1440/

[smoky oak]

there

[karmic sedge]

Thank you <3

[wispy ether]

https://tenor.com/view/backrooms-dancing-music-entity-entities-gif-25392627

[sterile temple]

activate.bat

[smoky oak]

alright, CMD is open there

[wispy ether]

means?

[smoky oak]

activated

[sterile temple]

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

[karmic sedge]

Good thing is I don't play at 1080p x)

[cyan snow]

*pip3 install

[wispy ether]

gta 4k

[smoky oak]

oh wow, that site doesn't even have the right GPU for me to calculate my bottleneck, wow lmao

[cyan snow]

pip install sometimes fuck up

[karmic sedge]

But I do play at 2k. Well, time for an 13th gen i9 I guess..

[sterile temple]

I created a folder somewhere, then moved a few recently installed extensions there so they were no longer in the extensions folder

[wispy ether]

wanna see bottleneck of my old laptop

[smoky oak]

its hella inaccurate anyways lol

[wispy ether]

where igpu bottlenecks its owqn cpu

[karmic sedge]

Sure

[karmic sedge]

Yeah it heavily depends on the game you play

[smoky oak]

of course an iGPU bottlenecks a CPU lmao

[sterile temple]

These are the extensions I moved out, then my install worked again

[smoky oak]

I can't even see what it would rate my bottleneck as it doesn't even have a standard 3080 lmao

[cyan snow]

does bottlenecks effect running AI?

[wispy ether]

if its fast ai, maybe

[smoky oak]

not really, just GPU as of now

[sterile temple]

pip3 I think is more for linux installs that have a python 2.7 install as the default python interpreter on hte system

[karmic sedge]

That's your laptop? Yeah I think that's not correct. 😅

[smoky oak]

running

[cyan snow]

does it effect SD? i have an older gen I9

[smoky oak]

bottleneck calcs are trash TBH

[wispy ether]

i just removed composable lora and ebsyntth

lets try

[smoky oak]

you'll be just fine

[wispy ether]

if you run sd on cpu

lmao

yes

but you should only run it if you dont have a gpu

[wispy ether]

still have the error

hmm

no module named pycuda

HMMMMMMMM

[karmic sedge]

I bottlenecked my old 3090TI with an i7 5820k. Even SD was wayyyy slower that it should have been. But I think it was slow because of the PCI-E bandwidth. Or am I wrong?

[smoky oak]

with how severe you have destroyed all your installations in the past, its no surprise lol

[cyan snow]

well, i don't. i have an i9 9900k or smthn and an RTX 4070TI

[wispy ether]

512x512

[smoky oak]

likely PCIE there

[smoky oak]

abnormally slow, even for a 3060

[wispy ether]

usually 7 tho

its like nothing changed

its the same speed

even a bit slower

[wispy ether]

what did you do

[karmic sedge]

xformers enabled?

[wispy ether]

no

opt thing

oh

what the fuck

[sterile temple]

must've been one of the flags or something in the launch file

[wispy ether]

my webui-user.bat is empty

wth

[karmic sedge]

well

[smoky oak]

FINALLY

I have 2.0.1

thank you, mikey

[sterile temple]

you might need to install the vcbuild tools

[cyan snow]

This is the speed i get on my 4070ti

[wispy ether]

where do i get them

[sterile temple]

https://aka.ms/vs/17/release/vc_redist.x64.exe

[karmic sedge]

512x512?

[wispy ether]

uh

[cyan snow]

yep

[wispy ether]

on a benchmark i saw 4070 getting 12it/s

[karmic sedge]

You should get at least ~18 it/s I think Oo

My 3090TI did 20

[wispy ether]

gonna restart pc to apply

[cyan snow]

then wtf is wrong?? i use cuda 11.8 and cudnn 8.6

im getting exactly half the it/s i should get

is it because im not on half precision?

[karmic sedge]

Uhm, good question. Where can you change the precision?

[wispy ether]

ok im back

[cyan snow]

in commandline args

if i recall correctly

[wispy ether]

in settings too i think

[karmic sedge]

fp32 will definitely use more resouces

[wispy ether]

almost no difference

[cyan snow]

are the benchmarks on fp16?

[karmic sedge]

rly?

[wispy ether]

@sterile temple

i installed the vs thing

[karmic sedge]

I don't know. I've never switched from 32 to 16 and vice versa

[wispy ether]

i have those

[cyan snow]

then why is it way slower then it should be?

[karmic sedge]

Is it this setting here?

[cyan snow]

no

it's in webui-user.bat

[wispy ether]

how do i check for update

[karmic sedge]

kk. I can't test it anymore because I am too lazy to plug the 3090TI back in. 😅
What's the command? I'll test it with the 4090 then.

[sterile temple]

do you have anaconda installed?

[wispy ether]

yes

[cyan snow]

Add --half-precision to command line arguments

[karmic sedge]

But half precision is fp16 and not 32

[wispy ether]

yes

fp16 is faster

and almost no difference in quality

[karmic sedge]

Yea he said he wasn't on half precision

[sterile temple]

@wispy ether go to stable-diffusion-webui\venv\Scripts folder

open terminal

[wispy ether]

ok

[sterile temple]

activate.bat

[wispy ether]

ok

[karmic sedge]

What are your system specs? Is your GPU in the uppermost slot and is it using it's full bandwidth?

[sterile temple]

conda install -c conda-forge pycuda

[wispy ether]

conda not recognized

[cyan snow]

yep

[sterile temple]

where is it installed?

[wispy ether]

wdym where

[karmic sedge]

Gen 5 m.2 or gen 4 in the gen 5 slot, if you have one? No idea how much pcie speed effects your it/s tbh

[sterile temple]

where is anaconda installed? the folder

[wispy ether]

idk

[sterile temple]

If conda ist not recognised, it's not in your path

[wispy ether]

ill reinstall it then

[sterile temple]

check your user folder first

[cyan snow]

it probably does, i actually bought a new m.2 like a few months ago, should be fine

[sterile temple]

or right click on one of those shortcuts and look at the path to the file

[wispy ether]

1sec i need to make space

for some reason only 100gb leftr

last time i checked it was like 400

[karmic sedge]

Yea but is it gen 4 or gen 5? Because some mainboards lower your pcie speed. Gen 5 uses more pcie lanes as far as I know.

[cyan snow]

gen 5, as much as im aware

[wispy ether]

🗿

[karmic sedge]

Do you have gpu-z installed?

[sterile temple]

you don't need to reinstall

[cyan snow]

wtf is gpu -z

[karmic sedge]

https://www.techpowerup.com/gpuz/

[wispy ether]

what do i do then

[sterile temple]

start menu, type env, then click on the option that comes up

[wispy ether]

ok

what file do i have to add to PATH

[wispy ether]

ok

[sterile temple]

double click on path

[wispy ether]

and then

[karmic sedge]

Oh and do you use the latest version of automatics?

[sterile temple]

you need to put the path to the anaconda3 folder in the list

[wispy ether]

where is the path

user?

[sterile temple]

yes your user folder

[cyan snow]

yes, even tho he is a bad person

ok, i got gpu-z

[sterile temple]

also add the anaconda3\Scripts folder

[wispy ether]

C:\Users\Lukag.conda

?

[sterile temple]

C:\Users\Lukag\anaconda3

[karmic sedge]

Run it and show us the main window

[sterile temple]

C:\Users\Lukag\anaconda3\scripts

[wispy ether]

ok

so both?

[sterile temple]

C:\Users\Lukag\anaconda3\library\bin

all 3

[sterile temple]

move them to the top

yep

[sterile temple]

and the C:\Users\Lukag\library\bin

[karmic sedge]

Oh you have the 12gb variant

[wispy ether]

now boot sd?

nvm we did something else

what were we doing

something in terminal

[cyan snow]

there is no other variant of 4070ti

[karmic sedge]

yea I am stupid

[sterile temple]

try running it, it might install pycuda otherwise we try install it manually

[wispy ether]

wasnt 4070 12gb and 4070ti 16gb?

[karmic sedge]

It was supposed to be the cheaper 4080

[cyan snow]

no. but still, the benchmarks says 20it/s, and im geatting about 10.

[wispy ether]

hm

[wispy ether]

use xformers

[cyan snow]

i am

[wispy ether]

use the opt attention command

i forgot

[cyan snow]

-opt-split-attention?

[wispy ether]

opt-spd-attention

[sterile temple]

@smoky oak how are you going?

[smoky oak]

got distracted 😅

[wispy ether]

https://tenor.com/view/distracted-gif-19280328

[smoky oak]

so, copy the tensor RT into the extensions tab?

[wispy ether]

follow the steps humblemikey told me if you havent yet

you need dev branch

takes forever

[smoky oak]

@sterile templeI should have a Tensor RT part of my SD now, right? Cause I can't find it in my UI

[sterile temple]

installed it?

[wispy ether]

i dont think its a TAB

or is it

[sterile temple]

it is

[wispy ether]

hm

for me there were no tab

[cyan snow]

holy shit, its doing 21.3it/s

[wispy ether]

and this is taking forever

Launching Web UI with arguments: --xformers
ControlNet v1.1.214
ControlNet v1.1.214
*** Error loading script: trt.py
Traceback (most recent call last):
File "C:\Users\Lukag\Desktop\stable-diffusion-webui\modules\scripts.py", line 274, in load_scripts
script_module = script_loading.load_module(scriptfile.path)
File "C:\Users\Lukag\Desktop\stable-diffusion-webui\modules\script_loading.py", line 10, in load_module
module_spec.loader.exec_module(module)
File "<frozen importlib._bootstrap_external>", line 883, in exec_module
File "<frozen importlib._bootstrap>", line 241, in _call_with_frames_removed
File "C:\Users\Lukag\Desktop\stable-diffusion-webui\extensions\stable-diffusion-webui-tensorrt\scripts\trt.py", line 11, in <module>
import pycuda.driver as cuda
ModuleNotFoundError: No module named 'pycuda'

[sterile temple]

did you extract that nvidia zip into the extension's folder?

[sterile temple]

close all terminal windows you have open, then open a new one in that venv\scripts folder

[karmic sedge]

Nice!

[sterile temple]

activate.bat

[wispy ether]

ok

[sterile temple]

conda install -c conda-forge pycuda

[smoky oak]

looks like this

[cyan snow]

im not sure if it affects quality, but yeah

[smoky oak]

is that how it should be?

[wispy ether]

(venv) C:\Users\Lukag\Desktop\stable-diffusion-webui\venv\Scripts>conda install -c conda-forge pycuda
Retrieving notices: ...working... done
Collecting package metadata (current_repodata.json): \

[sterile temple]

this is my folder

[karmic sedge]

I don't think so. I wouldn't have thought that this command makes such a huge difference tbh

[wispy ether]

https://tenor.com/view/kek-kekw-distorted-gif-19883544

[cyan snow]

yeah, more than x2 actually, it says in the wiki its enabled by default, no idea why it wasn't

[smoky oak]

@sterile templeummmm

my SD isn't starting-

[sterile temple]

did you change to the dev branch?

[wispy ether]

show erorr

[wispy ether]

.

[cyan snow]

without command

[smoky oak]

it won't go past installing requirements for web UI

[cyan snow]

with command

[wispy ether]

ok?

[smoky oak]

oh ok, it loaded finally, weird

[wispy ether]

the command doesnt do anything in quality

taking forever

[smoky oak]

guys, all it does is precompile the model

[wispy ether]

what

[smoky oak]

its the exact same stuff lol

[wispy ether]

ok

[smoky oak]

I did not

sorry, I don't think I was looking for that

[wispy ether]

i told you twice

[smoky oak]

do you mind if we switch to DM's @sterile temple ?

[wispy ether]

i just realized he blocked me

L

[cyan snow]

it always takes forever, it's making the enviroment, i think

[smoky oak]

I told you a long time ago I have you blocked, cause you like to spam the shit out of chat

its compiling the model

[wispy ether]

sytan thinks ur talking to him

cuz he blocked me

lol

im confused

[smoky oak]

no, I know he was talking to you. He thinks its slow cause its building, but its actually slow cause its compiling the model

[wispy ether]

lol va smearing

[sterile temple]

it can take a while

[wispy ether]

https://tenor.com/view/walter-white-walter-crying-walter-dying-gif-20964719

[smoky oak]

be warned that these tensor RT complies also take up a ton of room

[wispy ether]

space?

i need a new ssd jesus

i want a 2TB ssd only for python and another stuff like it

[smoky oak]

yeah, just 5 variations of the same model can use like 50 GB

[wispy ether]

since it takes alot space

An unexpected error has occurred. Conda has prepared the above report.

If submitted, this report will be used by core maintainers to improve
future releases of conda.
Would you like conda to send this report to the core maintainers? [y/N]: +

No report sent. To permanently opt-out, use

$ conda config --set report_errors false

[cyan snow]

nice language

[wispy ether]

@sterile temple

[sterile temple]

hmm

[wispy ether]

what now

[sterile temple]

pip install pycuda
does that work?

[wispy ether]

1sec

!!
check.warn(importable)
creating build\lib.win-amd64-cpython-310\pycuda\cuda
copying pycuda\cuda\pycuda-complex-impl.hpp -> build\lib.win-amd64-cpython-310\pycuda\cuda
copying pycuda\cuda\pycuda-complex.hpp -> build\lib.win-amd64-cpython-310\pycuda\cuda
copying pycuda\cuda\pycuda-helpers.hpp -> build\lib.win-amd64-cpython-310\pycuda\cuda
copying pycuda\sparse\pkt_build_cython.pyx -> build\lib.win-amd64-cpython-310\pycuda\sparse
running build_ext
building '_driver' extension
error: Microsoft Visual C++ 14.0 or greater is required. Get it with "Microsoft C++ Build Tools": https://visualstudio.microsoft.com/visual-cpp-build-tools/
[end of output]

note: This error originates from a subprocess, and is likely not a problem with pip.
ERROR: Failed building wheel for pycuda
Failed to build pycuda
ERROR: Could not build wheels for pycuda, which is required to install pyproject.toml-based projects

ill try downloading trhat

what do i need

note: This error originates from a subprocess, and is likely not a problem with pip.
ERROR: Failed building wheel for pycuda
Failed to build pycuda
ERROR: Could not build wheels for pycuda, which is required to install pyproject.toml-based projects

lets try

[smoky oak]

you're gonna break shit

[wispy ether]

i sent the wrong image lol

i meant this

should i install those

thats 9gb

jesus

8.36gb

[smoky oak]

bro, wait for him to respond before you break shit again, did you not learn from the kohya SS stuff? lmao

[wispy ether]

i didnt learn anything cuz i didnt get it fixed

atleast not really

[smoky oak]

I wonder why

lol

[sterile temple]

it might be easier to get a working version of conda installed

[wispy ether]

i actually do too

[sterile temple]

https://github.com/conda-forge/miniforge/releases/latest/download/Mambaforge-Windows-x86_64.exe

[sterile temple]

tick all

[sterile temple]

okay, leave the register... unticked

[wispy ether]

done

its installed

what now

[sterile temple]

do the start menu, type env, click the thing again

[wispy ether]

ok

what do i add

[sterile temple]

just show me the path list

[sterile temple]

is there a mambaforge folder in your user directory?

[wispy ether]

is --use-pytorch-cross-attention faster than xformers

[wispy ether]

yes

[sterile temple]

all those anaconda3 in the path, change them to mambaforge

[wispy ether]

@sterile temple

[sterile temple]

close all terminal windows

open a new one in the venv/scripts folder

mamba install -c conda-forge pycuda

[wispy ether]

launch.py: error: unrecognized arguments: --opt-spd-attention

uh

[wispy ether]

and now

[sterile temple]

looks like it installed pycuda. try running SD

[wispy ether]

*** !!
*** check.warn(importable)
*** creating build\lib.win-amd64-cpython-310\pycuda\cuda
*** copying pycuda\cuda\pycuda-complex-impl.hpp -> build\lib.win-amd64-cpython-310\pycuda\cuda
*** copying pycuda\cuda\pycuda-complex.hpp -> build\lib.win-amd64-cpython-310\pycuda\cuda
*** copying pycuda\cuda\pycuda-helpers.hpp -> build\lib.win-amd64-cpython-310\pycuda\cuda
*** copying pycuda\sparse\pkt_build_cython.pyx -> build\lib.win-amd64-cpython-310\pycuda\sparse
*** running build_ext
*** building '_driver' extension
*** error: Microsoft Visual C++ 14.0 or greater is required. Get it with "Microsoft C++ Build Tools": https://visualstudio.microsoft.com/visual-cpp-build-tools/
*** [end of output]

---

*** note: This error originates from a subprocess, and is likely not a problem with pip.
*** ERROR: Failed building wheel for pycuda
*** ERROR: Could not build wheels for pycuda, which is required to install pyproject.toml-based projects

---

Launching Web UI with arguments: --opt-spd-attention

launch.py: error: unrecognized arguments: --opt-spd-attention

---

Launching Web UI with arguments: --xformers
ControlNet v1.1.214
ControlNet v1.1.214
*** Error loading script: trt.py
Traceback (most recent call last):
File "C:\Users\Lukag\Desktop\stable-diffusion-webui\modules\scripts.py", line 274, in load_scripts
script_module = script_loading.load_module(scriptfile.path)
File "C:\Users\Lukag\Desktop\stable-diffusion-webui\modules\script_loading.py", line 10, in load_module
module_spec.loader.exec_module(module)
File "<frozen importlib._bootstrap_external>", line 883, in exec_module
File "<frozen importlib._bootstrap>", line 241, in _call_with_frames_removed
File "C:\Users\Lukag\Desktop\stable-diffusion-webui\extensions\stable-diffusion-webui-tensorrt\scripts\trt.py", line 11, in <module>
import pycuda.driver as cuda
ModuleNotFoundError: No module named 'pycuda'

[sterile temple]

https://stackoverflow.com/questions/64261546/how-to-solve-error-microsoft-visual-c-14-0-or-greater-is-required-when-inst

try following answer to question to install the build tools

you can try this

in the download folder where you downloaded the file

run this

vs_buildtools.exe --norestart --passive --downloadThenInstall --includeRecommended --add Microsoft.VisualStudio.Workload.NativeDesktop --add Microsoft.VisualStudio.Workload.VCTools --add Microsoft.VisualStudio.Workload.MSBuildTools

[wispy ether]

what file

[sterile temple]

the build tools install file

[wispy ether]

idk what file that is

[sterile temple]

https://aka.ms/vs/17/release/vs_BuildTools.exe

[wispy ether]

the exe?

[sterile temple]

vs_BuildTools.exe

[wispy ether]

how do i run a command with it

[sterile temple]

is it in your downloads folder?

[wispy ether]

yes

[sterile temple]

open a terminal in that folder

[wispy ether]

ok

i ran the command

any idea how to stabilize my wifi speed tho

max is 32mb/s

sometimes i just get only 8

or 10

[sterile temple]

move closer to the router? 😄

[wispy ether]

that doesnt change anything

if its a connection issue then it would be going up and down

but its constant 8mb/s

i think its a channel thing

but idk

my phone has no issues with wifi

only my pc

wifi card is Intel(R) Dual Band Wireless-AC 7265

ac

and 5ghz

and there is no driver

atleast for download

[sterile temple]

hmm, wifi does weird things, mine is much slower than what the modem can actually do

[wispy ether]

hm

maybe its the wifi card

idk

maybe i should upgrade anytime in future

to a newer and better one

also

the ms tool thing is installed

what now

@sterile temple

[sterile temple]

try running SD again? might have to restart your pc first?

[wispy ether]

https://tenor.com/view/kannafire-twinmomo29-kanna-chaos-gif-25407275

[sterile temple]

https://tenor.com/view/leslie-jordan-leslie-belly-dancing-belly-dance-dancing-gif-25526487

[wispy ether]

what do i do

[wispy ether]

@sterile temple

[sterile temple]

load the model you want to optimise

[wispy ether]

ok

the path?

or just the name

[sterile temple]

just how you normally do before generating images

[wispy ether]

there i use this

[sterile temple]

drop down at the top of hte screen

yes

select the model you want to optimise

[wispy ether]

i did

but i mean what do i put there

[sterile temple]

leave blank

[wispy ether]

is --opt-spd-attention right tho

or is it wrong typed

[wispy ether]

and then this?

[cyan snow]

idk, whatever that command is it doubled my performance, i don't think it's fp16 because fp16 is noticable and this is not

[wispy ether]

its always fp16 if you dont use --no-half and or --precision full

[cyan snow]

i use no half, and when using opt-spd-attention, it doubles performance with no change

[sterile temple]

you need to find the file it made first, and copy the full path to that file and put it in the box at the top

see how mine has the full path to the onnx file

[wispy ether]

C:\Users\Lukag\Desktop\stable-diffusion-webui\models\Unet-onnx

[cyan snow]

onnx increases performance? i thought diffusers is the most efficient optimization

[wispy ether]

tensorrt

[cyan snow]

never heard of that

[wispy ether]

🗿

https://developer.nvidia.com/tensorrt-getting-started

[wispy ether]

@sterile temple cannot find the path

error

FileNotFoundError: [WinError 3] Das System kann den angegebenen Pfad nicht finden: ''

---

FileNotFoundError: [WinError 3] The system cannot find the path specified: ''

[sterile temple]

check the path again

[wispy ether]

its the right one

wait

now after i removed output name it did something

\src\google\protobuf\io\coded_stream.cc:604] Reading dangerously large protocol message. If the message turns out to be larger than 2147483647 bytes, parsing will be halted for security reasons. To increase the limit (or to disable these warnings), see CodedInputStream::SetTotalBytesLimit() in google/protobuf/io/coded_stream.h.
[libprotobuf WARNING **************************************************************************\externals\protobuf\3.0.0\src\google\protobuf\io\coded_stream.cc:81] The total number of bytes read was 1721624086
[06/03/2023-14:44:03] [W] [TRT] onnx2trt_utils.cpp:374: Your ONNX model has been generated with INT64 weights, while TensorRT does not natively support INT64. Attempting to cast down to INT32.
[06/03/2023-14:44:04] [I] Finished parsing network model. Parse time: 2.19733
[06/03/2023-14:44:04] [I] [TRT] Graph optimization time: 0.399335 seconds.
[06/03/2023-14:44:04] [I] [TRT] [MemUsageChange] Init cuBLAS/cuBLASLt: CPU +0, GPU +8, now: CPU 21384, GPU 2476 (MiB)
[06/03/2023-14:44:04] [I] [TRT] [MemUsageChange] Init cuDNN: CPU +0, GPU +8, now: CPU 21384, GPU 2484 (MiB)
[06/03/2023-14:44:04] [W] [TRT] TensorRT was linked against cuDNN 8.9.0 but loaded cuDNN 8.7.0
[06/03/2023-14:44:04] [I] [TRT] Local timing cache in use. Profiling results in this builder pass will not be stored.

[cyan snow]

is tensorrt 8.6EA or 8.6GA better?

[wispy ether]

idk

its stuck

[cyan snow]

how do i deploy it @wispy ether

[cyan snow]

got the zip

[wispy ether]

read through the whole convo where humblemikey explained it to me

you need dev branch

[wispy ether]

extension folder

but unzip

[sterile temple]

looks like it's running

takes about 15 minutes

[wispy ether]

more like running forever

oh

https://tenor.com/view/emoji-died-kek-plus-gif-19898158

[wispy nest]

4k

[wispy ether]

4k is 3840x2160

@sterile templedo i have to put the file into the models folder of sd

the .trt one

[wispy nest]

its even bigger than 4k

[wispy ether]

why not if i download it

[wispy ether]

bruh

how did you even

[wispy nest]

click open in browser

[wispy ether]

stable diffusion doesnt generate good in higher resolution

even with hires fix

[wispy nest]

yes it does

with ultimate SD Upscale + controlnet tile

[wispy ether]

bruh

[wispy nest]

also inpainting a lot

took me ~1 hour to make that image

[sterile temple]

it should just save to the correct folder

[wispy ether]

yea but how do i use it then

[wispy nest]

https://github.com/Coyote-A/ultimate-upscale-for-automatic1111

https://github.com/Mikubill/sd-webui-controlnet

[sterile temple]

it will use it if you use the model with the matching name

[wispy ether]

ok

[wispy nest]

this is what i used to get the high res

[wispy ether]

no speed difference

[cyan snow]

same, it does just as good as diffusers

[wispy ether]

@sterile templedo i have to do something else? cuz the speed is the same

terminal says its fp32 because tensorrt doesnt use int64

or int32

idk

i want fp16

[cyan snow]

im still on around 21it/s with this

[sterile temple]

is it showing messages like these when you run a gen for the first time?

[sterile temple]

I'm currently trying to figure out why mine is running slow again. It was running faster before

[wispy nest]

imagine having a gpu with tensor cores

[wispy ether]

imagine having amd

https://tenor.com/view/yeah-i-flip-gp-us-gpu-graphics-card-scalper-scalping-graphics-card-gif-24988488

[wispy nest]

according to benchmarks the 7900 xtx isnt too bad

its like a 3090

same price even

[wispy nest]

but holy power consumption

and weight

this thing is literally a brick

[wispy ether]

@sterile templedm

[wispy nest]

mfw i get 2 it/s

or 1

[karmic sedge]

[wispy nest]

what gpu do you have

4090 i assume

[karmic sedge]

Yeah and I use TensorRT

[cyan snow]

I got 4070ti and i get 21it/s

[wispy nest]

crazy numbers

[cyan snow]

does it make a difference for you?

[wispy nest]

imma wait for a new platform to emerge

[smoky oak]

I got tensor RT working for me

[wispy nest]

because these gpu prices are ridiculous

[smoky oak]

made a decent difference on my 3080

[karmic sedge]

From ~33 to ~55 it/s

[smoky oak]

I went from 15.2it/ to 27it/s

[karmic sedge]

Sweet

[wispy ether]

https://tenor.com/view/crying-emoji-dies-gif-21956120

i get 2it/s on 768x768

[karmic sedge]

I am excited to see what happens when Nvidia releases their version for webui

[wispy ether]

tensorrt

[smoky oak]

I am getting 11.9it/s at 768x768

[smoky oak]

same!

[karmic sedge]

^Also with TensorRT

[smoky oak]

this is soooo much faster lmao

[karmic sedge]

Yeah 😄

[smoky oak]

now if only I can get it working with batch size

[karmic sedge]

And more than 75 tokens

[smoky oak]

what?

mine is at 375 tokens 😅

[wispy ether]

for me and @sterile temple its slower

tf

wdym tokens

[smoky oak]

mine is min and max 512x768 for both dimensions, and 375 max tokens

[karmic sedge]

It doesn't work for me if I use more then 75 tokens. No idea what I am doing wrong D:

[wispy ether]

what does tokens do

[smoky oak]

hmmm, not sure

[karmic sedge]

Yea I use the same resolution

[smoky oak]

how many words your prompt can have

[karmic sedge]

Did you download TensorRT EA or GA?

[wispy ether]

i almost never use more than 20 words

[smoky oak]

if I can get 512x768 both directions with BS 8, I am sold

[wispy ether]

ga

[smoky oak]

wow lmao

[wild sorrel]

11 it
on my 1050ti on 768x768

[karmic sedge]

Yea I am using EA, thats the problem I guess

[wispy ether]

liar

[sterile temple]

@smoky oak @wispy ether I gotta head to bed guys. Hopefully I can get it going faster tomorrow

[smoky oak]

ummmm.... lol

[wispy ether]

when is tomorrow

timezones exist

[smoky oak]

sounds good! thanks again for the help man

[wispy ether]

YOU LIAR

[sterile temple]

I'm in Australia

[sterile temple]

np

[wispy ether]

1050ti doesnt even have tensor cores

[wild sorrel]

How exactly are we counting? 1 image \ full batch \ sampler?

[wild sorrel]

yep

[smoky oak]

it/s

[wispy ether]

you said 768x768

[wild sorrel]

yea

[wispy ether]

i get 2it/s on 3060

11it/s doesnt make any sense on a 1050ti

[smoky oak]

its not possible lol

[wispy ether]

fr

well.......

maybe if you overclock like 20ghz

[smoky oak]

oh shittttt

welp, found a limit of tensor RT

[wispy ether]

tell us

[smoky oak]

no high res fix

[wispy ether]

ok

[smoky oak]

that is a huge hit for me, damn

[karmic sedge]

Almost

[smoky oak]

schpeed

[sterile temple]

without trt 8 and 5 with, so need to find a way to fix. anyway off to bed

[wild sorrel]

Just tell me how are we testing =.=
What batch size , what sampler

[wispy ether]

same

waoit

how 8

i get 7

we have same gpu

[karmic sedge]

[smoky oak]

512x512 takes 1.09s on my GPU now 😅

[smoky oak]

that is WITH post processing lol

[wispy ether]

how to enable it

[wild sorrel]

or don't , I'll go watch youtube instead

[wispy nest]

cheating

[smoky oak]

we are talking about iterations per second, no matter what sampler, a 1050ti isn'y doing 11it/s lol

[wispy nest]

TensorRT only works with 20XX and above right

[smoky oak]

30xx and above I think

[wild sorrel]

I was doing single image run, didn't want to waste too much time on it.
I guess you all did full batch and looked 1out 8 images speed?

[karmic sedge]

Uhm, does someone here have a link for TensorRT 8.1.6.1 GA?

[wispy ether]

rtx

[wispy nest]

[smoky oak]

just a sec

[wispy nest]

my GTX

[wispy ether]

tensorrt makes it look shitty

[smoky oak]

https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/secure/8.6.1/zip/TensorRT-8.6.1.6.Windows10.x86_64.cuda-11.8.zip

[smoky oak]

literally doesn't change image quality at all, but ok lol

[karmic sedge]

Thank you <3

[smoky oak]

no worries

[wispy ether]

https://tenor.com/view/man-ricky-berwick-oh-dear-oh-my-god-oh-man-gif-26263798

[wispy nest]

@wispy ether https://civitai.com/models/47085/envybetterhands-locon

[wispy ether]

i tried that already

it doesnt do an ything

[wispy nest]

it works for me

[smoky oak]

my images look just as good as they did before, tensor RT doesn't change image quality at all

[wispy ether]

tensorrt doesnt do anything for me

[wild sorrel]

prompting issue

[wispy ether]

just slower

[smoky oak]

certifiably lol

damn this is fast lmao

[wispy ether]

what

everytime after i use the model 1 time that happens

then i have to make a new one

[smoky oak]

I just need this thing to work with BS 8 :<

or wait

[wild sorrel]

disposable model

[wispy ether]

imma uninstall tensorrt .-.

[smoky oak]

it might actually be faster to do batchcount 8 now lol

[wispy ether]

size*

what even does that mean

raise Exception(f'bad shape for TensorRT input {binding}: {tuple(shape)}')
Exception: bad shape for TensorRT input x: (8, 4, 64, 64)

---

[smoky oak]

don't correct me, you're wrong lol

I said what I meant lol

[wispy ether]

yro´ue

batch count is just to make more images

instead of clicking generate everytime

[smoky oak]

I know what batch count is

[wispy ether]

so

how does it make it faster

how do i quit dev branch

[wild sorrel]

I want your model and prompt
Throwing it into negative didn't do the thing on my model

[smoky oak]

oh wait!

BS 2 seems to be working

[wispy ether]

how do i quit dev branch

[smoky oak]

¯_(ツ)_/¯

good luck

[wispy ether]

?

[wild sorrel]

I mean...you change "branch_name" to actual branch name - usually it's "master"

Edit: edited to avoid further confusion.

[wispy ether]

main???

[wild sorrel]

master , right 😄

[wispy ether]

????

[wild sorrel]

git checkout master

[wispy ether]

ah

[wild sorrel]

That's branches names in a1111

[karmic sedge]

@smoky oak
Do you get the same error message when you convert onnx to tensorRT?
[06/03/2023-15:41:27] [TRT] [W] TensorRT was linked against cuDNN 8.9.0 but loaded cuDNN 8.7.0

[wispy ether]

just dont use it

its kinda broken

without tensorrt

with tensorrt

[karmic sedge]

Is there another way?

[smoky oak]

No

For me, I get some form of out of range error

[karmic sedge]

Uhm you have a 20xx series card, or?

[wispy ether]

30

[smoky oak]

He has a 3060

Of which I am wondering the issue stems from the cucked VRAM bus

[karmic sedge]

Okay weird.

[smoky oak]

What GPU do you have? @karmic sedge

[karmic sedge]

4090

[smoky oak]

Ah

It would be wayyyy faster for you then haha

[karmic sedge]

Yea I get almost 60 it/s

[smoky oak]

Oh my bad, I didn't realize you were the same guy haha

[karmic sedge]

Damn my english is broken today

Heh no problem 😄

[wispy ether]

@karmic sedgewhat tensorrt do u have

[smoky oak]

I am on mobile now, so I can actually see profile pics now haha

[wispy ether]

what did u do

mine is slower

[karmic sedge]

8.1.6.1

[wispy ether]

i mean ea or ga

[wild sorrel]

spinny things confuse me , can't diffirintiate people anymore

[karmic sedge]

GA now

[smoky oak]

Please don't bother with nep, he's bound to not listen to anything you help him with lol

[smoky oak]

Yeah haha

[wispy ether]

you didnt even try it#

[karmic sedge]

Good to know

[wispy ether]

do you believe everything someone tells you

[karmic sedge]

No, just if it's Sytan

[wispy ether]

and sythan is

who

[smoky oak]

Took me and 6 other people 9 hours of helping him to install kohya, and it STILL didn't work lol

[karmic sedge]

A random stranger on the internet I guess

[wispy ether]

even with a youtube tutorial it didnt

[smoky oak]

Probably cause you did like 30 things we said not to do lol

Anyways, I don't wanna drag this on

[karmic sedge]

you did the mem attention thing, right?

[wispy ether]

like what

[smoky oak]

I am using all 4 optimizations in my bat

[karmic sedge]

All 4? Then remove no mem attention Oo

[smoky oak]

SDP, SDP no memory, is last, and quad

I found all 4 together were the fastest for me

[wispy ether]

which 4

[karmic sedge]

Srsly?

[wispy ether]

i only know 2

[smoky oak]

Yeah, for me at least

[karmic sedge]

I mean, doesn't SDP and SDP no mem cancel each other out?

[cyan snow]

yo, why does the TensorRT tab not show up on my webUI

[smoky oak]

I don't believe so, I believe it just makes it a little less memory efficient to be a little faster

[wispy ether]

cuz no dev branch

[karmic sedge]

^

[smoky oak]

You have to go dev branch

[wispy ether]

@smoky oakcan you send ur args command line

[smoky oak]

I am not qualified enough to explain how lol

[cyan snow]

i did git checkout dev

[smoky oak]

Not right now

Maybe later

[wispy ether]

🗿

[karmic sedge]

Then do git pull next

[wispy ether]

where can i see all commands

[cyan snow]

git pull dev?

[wispy ether]

what even is the trigger word

[karmic sedge]

Just git pull after you switched to the dev branch

[wispy nest]

its in the description somewhere

its 500 mb tho, better just use it with inpainting to fix stuff

[wispy ether]

i already have it tho

[cyan snow]

ok its doing smthn

[karmic sedge]

Oh and make sure you have TensorRT 8.1.6.1 extracted in the extension folder

[wild sorrel]

git checkout dev
git pull

[cyan snow]

um, i got tensorrt 8.6.1.6

[karmic sedge]

kk

[wispy nest]

inpaint to fix

[wispy ether]

idk how to inpaint

[wispy nest]

whole image, 0.75, fill

send to inpaint button

just draw over the hand

by clicking on the image on the left

[wispy ether]

full hand or just fingers

[wispy nest]

whole thing

[cyan snow]

still doesn't show up

[wispy ether]

is that right

[wispy nest]

yeah, now just do 0.75 denoise strength, either fill or latent noise, euler A 30 steps

[wispy nest]

and then let it spam some images

[smoky oak]

||@wispy nest be prepared for him to blame you when it doesn't work||

[wispy ether]

:I

[wispy nest]

[smoky oak]

Lmfao

[wild sorrel]

yea but you can give it a little bit more space around fingers to work with, so it has more space to manipulate

[wispy nest]

yeah that works aswell

just make sure that it paints the whole image or it will add little people in your masked area

inpainting is a tool you should practice to use because its insane to do high res

[wispy ether]

with batch count?

[karmic sedge]

does your .bat look like this?

[wispy nest]

as many as you want

[karmic sedge]

Just to be on the safe side

[wild sorrel]

and make prompt related only to area you are inpaiting

[wispy ether]

what does batch size do on generating tho

[wispy nest]

i just do 100 and click interrupt when i got something nice

[cyan snow]

yes..?

[wispy nest]

batch size is how many images at once

[wispy ether]

even worse

[karmic sedge]

Weird

[wispy nest]

you did not select fill

or latent noise

you selected original

select either fill or latent noise

original will just make it denoise the original bad shapes, fill will add some blurry mess and latent noise will force it to redraw entirely

[cyan snow]

im also now getting error: error: command 'C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Tools\MSVC\14.29.30133\bin\HostX86\x64\cl.exe' failed with exit code 2

[karmic sedge]

Oh wait

[wispy nest]

right hand looks ok now

just repeat it with the left hand

[wispy ether]

shit is cursed

[wild sorrel]

I think that's just either not a good model or a not a good prompt...anime models are usually somewhat decent with hands

[wispy nest]

you can also just describe the hand in the prompt

[wispy ether]

5 fingers, 1 hand

🏰

[wispy nest]

'left hand with palm facing upwards, stretched out fingers, thumb'

[cyan snow]

does anyone know what this means?

[wispy ether]

[oak osprey]

@strong sonnet they charge for network bandwidth on vast.ai too wtf lol

[wispy ether]

is that even right

[karmic sedge]

Install visual studio 2019 build tools like this

[wispy nest]

yes

whats your prompt

[karmic sedge]

And delete the other versions if possible

[wispy ether]

4

words

bruh

[karmic sedge]

Also make sure to install nvidia cuda toolkit@cyan snow

[wispy ether]

its automatically installed with nvidia driver

[wispy nest]

prompt

[karmic sedge]

I had to do it manually

[wispy ether]

girl, angel, fire storm lyco:GoodHands-beta2:1.0

[karmic sedge]

No idea why

[wispy ether]

the lyco thing doesnt do anyting

[wispy nest]

does it say anything inside your terminal

also whats your prompt

[wispy ether]

....

Apply LyCORIS model: GoodHands-beta2
Text encoder weight: 1.0
Unet weight: 1.0
DyLoRA Dim: None

[cyan snow]

bruv

[cyan snow]

the option you checked doesn't show up

[karmic sedge]

desktop development with c++

[wispy nest]

do this

[wild sorrel]

wtf lol

[wispy ether]

no

too big

too much difference

where can i see all arg commands

[wispy nest]

no just the nice hands, perfect hands and the polydactyl thing