City

It seems to run into the following compile error for me on my Pascal-based card (Tesla P40 with Compute 6.1) on Ubuntu 22.04.2 LTS: ``` GPTQ-for-LLaMa/src/gptq_llama/quant_cuda/quant_cuda_kernel.cu(977): error: no instance of...

As far as I can tell that code is already present in `/src/gptq_llama/quant_cuda/quant_cuda_kernel.cu`. Removing the `#ifdef` so it always runs also doesn't change anything. I'm not too well versed in...

It seems to be working, thank you! @iwalton3

Interesting. I've briefly looked at 3 before but it needed some other flavor of T5 that looked annoying to implement. This one just seems to use CLIP though. I'll see...

These models only work (well) on the resolutions they were trained on as far as I can tell. You can try to use one of the tiled sampling solutions like...

The PixArt team only released one controlnet so far as far as I know, and those conditioning nodes are meant to be used with that one (see issue #13 )....

Thanks for the heads up, I pushed a quick fix. Could definitely be improved but I think SD3 will (in part) use T5 as well, so I'm hoping to reuse...

As a follow up on this, I did some testing and trained a VAE by setting the encoder to constant (copying the weights from the v0.9 VAE). This let me...

>What hardware x time would be required for a high quality model? Good question. I don't have much experience with distributed/large scale training. I just used a slightly-modified version of...

IIRC the UNET did a pretty good job at getting rid of the weird artifacts. It forces the latent back into a format the VAE actually expects, so there usually...