Closed pwalski closed 1 month ago
EDIT: OUTDATED
Error response returned by Automatic API
recommends using --disable-nan-check
and --no-half
parameters.
It happens because there were some problems with model I used.
ya-runtime-ai
provides a way to configure --ckpt
cmdline param using golem.srv.comp.ai.model
property.
There is no other way to add additional cmdline parameters, and adding some of them might be necessary to work with some models.
Maybe we should have a property where user could add additional cmdline parameters? This list would be validated using a whitelist of allowed parameters (like --no-half
, --no-half-vae
, --medvram
, --disable-nan-check
, etc. More of them here https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Command-Line-Arguments-and-Settings)
It seems like a property specific to Automatic
runtime, but vllm
might also need an additional --chat-template
cmdline arg, so adding some general extra runtime cmdline args property could be useful later, and only a validation of this property (checking against whitelist) would be a runtime specific implementation.
Issues:
500
error response as200
.I tried to use example
Golem.Tools/App
with model https://huggingface.co/stabilityai/stable-diffusion-2-1/resolve/main/v2-1_768-ema-pruned.safetensors?download=true?download=true Runtime started correctly, model got loaded (loading incompatible model sometimes fails), but requests are failing, and logs print recommendations about adding some config options to Automatic commandline which user is unable to do:User received following response
200 OK
:It seems that Automatic API returned
500
, but gsb http proxy returns200
.Versions:
(SHA256 B6F47B478BE36118D3DB732173C65C194BFF64342E28BFEC776F182A37DA69A4) (ya-runtime-ai still prints 0.1.1)
Originally posted by @pwalski in https://github.com/golemfactory/ya-runtime-ai/issues/79#issuecomment-2060920089