fast-stable-diffusion
fast-stable-diffusion copied to clipboard
why not using kaggle instead of colab? There is P100, T4X2 for 36 hours in a week for free
Am I missing something?
I must agree with this one. I am using kaggle for lots of works.. and it is a viable solution. the 2xT4 actually improves the generation by a lot comparing to colab .... also... there is a FREE P100
Working on it, as soon as I'm done with new features in Dreambooth
issue with kaggle is mounting the g-drive or equivalent...
thats a big issue, but you can register multiple accounts and run at the same time so thats good, how do you guys download big files from it ?
you dont need to mount. You can share a model on your gdrive and then just use !Gdown --id <shared_link_id> and it downloads to your working area. The generated files can be zipped and download... easy peasy! Worth the trouble for the extra computational power...
The main differences is that all /content directories must be replaced by /kaggle/working .. and there are some details with gradio and streamlite .. (although they do work.. just not a direct port from colab to kaggle)
Has anyone got Kaggle working with fast-dreambooth? I'm getting an error when trying to run the training cell: OSError: /opt/conda/lib/python3.7/site-packages/torch/lib/../../nvidia/cublas/lib/libcublas.so.11: undefined symbol: cublasLtHSHMatmulAlgoInit, version libcublasLt.so.11
Has anyone got Kaggle working with fast-dreambooth? I'm getting an error when trying to run the training cell:
OSError: /opt/conda/lib/python3.7/site-packages/torch/lib/../../nvidia/cublas/lib/libcublas.so.11: undefined symbol: cublasLtHSHMatmulAlgoInit, version libcublasLt.so.11
Same here.