Disty0

I suspect this is an issue with a common Intel library. Because this same issue happens on IPEX as well: https://github.com/intel/intel-extension-for-pytorch/issues/476

You can run separate taggers at the same time. You will start to get CPU bottlenecked to 60~ images per second per tagger. Run one part of the dataset on...

Another workaround that doesn't require code change on diffusers side: ```py backup_clip_txt_pooled_mapper = copy.deepcopy(sd_model.prior_pipe.prior.clip_txt_pooled_mapper) pipe.prior_prior = pipe.prior_pipe.prior = nncf_compress_model(pipe.prior_pipe.prior) pipe.prior_prior.clip_txt_pooled_mapper = pipe.prior_pipe.prior.clip_txt_pooled_mapper = backup_clip_txt_pooled_mapper ```

Some notes about NNCF: - It is autocast. Meaning model weight will be stored using INT8 and will be casted to the original dtype (BF16/FP16 etc) on forward pass. -...

This is how i implemented it in SDNext: https://github.com/vladmandic/automatic/blob/master/modules/sd_models_compile.py#L104 I am treating it as if it was compiled because it breaks loading Loras, not because it is compiled. (It's not...

Can you try Euler a to see if it's an issue with UniPC / Linalg Solve?

Can't reproduce any of this in either native install or my docker image. "MetadataIncompleteBuffer" means corrupted model. Run sha256sum on the model file and check if it matches with the...

If you can pinpoint which packages are missing, i can. You've installed 20GB~ packages that will never be used.

https://github.com/vladmandic/automatic/wiki/Intel-ARC#errors-with-ipexrun-on-linux

This seems like the same issue as IPEX 2.1 + MKL / DPCPP 2024 issues on IPEX side. UniPC and original backend is known to crash. I don't plan to...