Wagner Bruna
Wagner Bruna
@stduhpf , would you please rebase this onto master? There are a few conflicts due to the VAE changes for Wan support.
> I haven't tested it, I know it builds fine, but I'm not sure if it actually works (I can't test it right now) No problem, I'll test it here.
@leejet , a quick question: the VAE tiling doesn't really change the loaded models, and its usefulness depends on the chosen image resolution (we even have a workaround on Koboldcpp...
> @wbruna do you want to do these changes yourself or should I go ahead and start working on it? I won't be able to until Friday, so feel free...
master-fce6afc is working for me (gfx1102, ROCm 6.4.1). Which model are you running? And with which command line parameters?
> `sd -m '/[modeldir]/sd_xl_base_1.0.safetensors' --vae '/[modeldir]/sdxl_vae.safetensors' --t5xxl '/[modeldir]/t5xxl_fp16.safetensors' -H 1024 -W 1024 -p 'a lovely cat holding a sign says \"Stable diffusion XL\"' -v --cfg-scale 4.5` Not likely to be...
Yeah, I agree it should work. I'd suggest asking around on the [llama.cpp](https://github.com/ggml-org/llama.cpp) project; you'd have much more visibility there (there is a separate ggml repository, but most ggml development...
There is a known issue that could cause a GET_ROWS error (#837), but not for f16 models... maybe that sd_xl_base_1.0.safetensors of yours has an uncommon quant? Does it work with...
Perhaps we could have a 'staging' branch too, for more involved changes? That way, additional functionality could be pre-merged and better tested without disrupting the releases. There is a lot...
There is "stable", and there is "bugs are actually getting fixed there" 🙂 As much as I want the new features myself, IMHO it's best to set up the new...