Some weight files are duplicate

#2
by zhy2050 - opened

Those file are exactly the same. I don't get why.
MD5, file_path
597c24855313598cb26066d28006596d ./text_encoder/model-00001-of-00002.safetensors
597c24855313598cb26066d28006596d ./text_encoder/model.bf16-00001-of-00002.safetensors

b16329d04bd0ce6f2a524e5bf8ef5908 ./vae/diffusion_pytorch_model.bf16.safetensors
b16329d04bd0ce6f2a524e5bf8ef5908 ./vae/diffusion_pytorch_model.safetensors

bf07c46c7021d3def62bfad765af1222 ./text_encoder/model-00002-of-00002.safetensors
bf07c46c7021d3def62bfad765af1222 ./text_encoder/model.bf16-00002-of-00002.safetensors

It is because of something to do with how the pipeline.from_pretrained(variant = "bf16") works. When the variant is specified, all the safetensors files being loaded must have the variant string in the name. If you just run the provided code as is, only the files corresponding to the specified variant will be downloaded and used.

Sign up or log in to comment