runpod-worker-a1111 and loras
I dont think my loras are working with this worker? But it seems to be able to get loras with the /sdapi/va/loras
https://github.com/ashleykleynhans/runpod-worker-a1111/blob/main/docs/api/a1111/get-loras.md
so am i able to use loras with this worker or no?
GitHub
runpod-worker-a1111/docs/api/a1111/get-loras.md at main · ashleykle...
RunPod Serverless Worker for the Automatic1111 Stable Diffusion API - ashleykleynhans/runpod-worker-a1111
76 Replies
actually you know what i did the GET /sdapi/va/loras
and the output came back empty? i think i might have to put the loras in the wrong place?
they are currently in the
/workspace/stable-diffusion-webui/models/Lora# V1?
Is that the right endpoint payload
Yeah it should work
You must have put the LoRA in the wrong folder if its empty
Maybe you also need to call the refresh checkpoints endpoint
this is the folder
im looking at this https://www.generativelabs.co/post/getting-loras-to-work-with-a1111-api
why is lora lowercase
Linux is case sensitive
i just changed it lowercase
it was uppercase before
should it be uppercase?
Leave it the way it was before, it won't work if you change it because Linux is case sensitive
I think it's that way for a1
Oh nvm I don't remember it so well
it was always uppercase but since it dont work i changed it but
i changed it back to uppercase
anyways
let me do a refresh
thats how it should be
yeah it didnt detect it
btw im on serverless endpoint
Sorry not refresh checkpoints, thats for models
GitHub
runpod-worker-a1111/docs/api/a1111/refresh-loras.md at main · ashle...
RunPod Serverless Worker for the Automatic1111 Stable Diffusion API - ashleykleynhans/runpod-worker-a1111
refresh-loras
yes
/sdapi/v1/refresh-loras
{
"delayTime": 463,
"error": "A1111 status code: 405",
"executionTime": 84,
"id": "c95c0ac3-86e1-4c22-b6da-07dbff387401-u1",
"output": {
"output": {
"detail": "Method Not Allowed"
}
},
"status": "FAILED"
}What payload did you use?
nvm lol i did get lmao hold on
waiting in queue
btw i get coldstarts alot even when im like
constantly generating
the GPUs seems to get taken away from my serverless
randomly
Did you enable flash boot on your endpoint?
yes
{
"delayTime": 63844,
"executionTime": 60,
"id": "00330fa9-ac79-4a2d-aa10-3df3a5eede64-u1",
"output": [],
"status": "COMPLETED"
}
its still not detecting my loras...do i have to name them with an extension? like add-detail-xl.tensorflow or somethingnot .tensorflow, .safetensors
but yes, they should typically have an extension like .safetensors, .ckpt etc
also is there a way to refresh the serverless endpoint with the new data in the network volume? i find myself constantly deleting the serverless and remaking it when i make changes to the network volume
You shouldn't have to delete it and recreate it, it should update most things automatically, you may just need to call the refresh endpoints
ohh right
silly me
ok that worked lol the loras are there
i just didnt have the extensions for them lmao
thx ❤️
Great stuff
Huh u can use it from the prompt
i didnt name the lora files with the .safetensors extension in the folder so the GET lora endpoint wasnt picking them up
they definitely working now
Ooh
Okok files ext
Yeah all model types need a file extension or they won't be found, whether its SD, ControlNet, LoRA etc
yeah i knew that too cuz i had the same problem with models and then i renamed them and they was found
Probably to do with the way models are loaded, for example ckpt files can have malware embedded into them (which is why people switched to using .safetensors instead) so need to know the filename to determine how to load it, because if the extension is .ckpt, safe unpickle can be used to avoid malware from being loaded into the system.
I typically try to avoid .ckpt models and use .safetensors only
yeah i havnt came across any ckpt yet
Most people switched away from it luckily
do you guys know if there is an A1111 API schema somewhere? like a complete one? for example i found out that i need to set Emphasis: No norm
but dunno what the API setting is and its not listed in the ashley wiki
woop nvm i found a pdf
That one might not be up to date
it doesnt list it anyways
I forgot wheres the a1111 docs is
yeah its from 2023
When you run one in like pods, it has one endpoint for the docs as html
yeah
Like from the webui
leme try one
just set it up as the pod
Right
RunPod Fast Stable Diffusion
runpod/stable-diffusion:fast-stable-diffusion-2.4.0
template
Not sure whether that one has
--api enabled
If --api is not enabled, you can't access the API docsi found it
its "emphasis": "Original",
POST
/sdapi/v1/options
Set Config
inside there
What's that for hahah
so i spun up a pod and then just accessed the /docs
Never heard of it
Right click, save as html so you don't have to memorize it lol
i dont even know
all i know is it stops pony xl from crashing
not crashing but
there are so many API config options
lol
Oh is that a1111
Oh the Emphasize normalization yeah yeah
yeah its A1111 the option in the webui
nice
now i have all the APis lol
Hahah yeah
Gonna be reading that for a day? Lol
Alot of options
yeah read all and then i wont have any issues when ppl talk about some setting im like oh im on API i dunno if i can do it lol
do you guys know why everytime i make a request its a cold start? and it takes like 4-5 minutes per generation
2024-06-14 16:17:57.805
[sjvl6zqmehakh0]
[info]
Service not ready yet. Retrying...
its like everytime im done gen the gpu gets taken away
Both in the same like minute?
yeah
once its done i usually send another request immediately right after
it is low availability
right now
so i dunno
i usually pick 2 but today im picking all of them
except the 80s
Oh yeah maybe try again
Did you have refresh worker?
Cold starts are random
wat is refresh worker
Oh nvm
then its just random, flashboot isnt activated all the time, only if you have continous requests i think
yeah its random sometimes its fast
even after some time in between
im not sure what happened so maybe the flashboot wasn't activated or maybe your worker got throttled because of the low availibility
but flashboot is kinda random for low amount request ig
pretty sure its low availability, im considering copying pasting the network volume to a different timezone but they dont seem to have much availability either from the preview, are you guys gonna add more gpu lol
ofc runpod will eventually
unless something happened