Hello,
I tried to contact you by mail for your Replicability Stamp submission, but didn't get a response. Just in case, I'll copy-paste the issues I had below.
"1/ When downloading the required models using your script (models/download_hf_models.py), I had an error for the first model v1-5-pruned.ckpt:
Cannot access gated repo for url[ https://huggingface.co/benjamin-paine/stable-diffusion-v1-5/resolve/26a823710f75136819d791422b0b8686afbe784b/v1-5-pruned.ckpt](https://huggingface.co/benjamin-paine/stable-diffusion-v1-5/resolve/26a823710f75136819d791422b0b8686afbe784b/v1-5-pruned.ckpt).
Access to model benjamin-paine/stable-diffusion-v1-5 is restricted. You must have access to it and be authenticated to access it. Please log in.)
As a work-around, I downloaded this file manually from this link (it looked like the same model):
https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5/blob/main/v1-5-pruned.ckpt
Do you know if this permission issue is an error on my side ? If not, could you either change the script, or add instructions on how to download this file manually ?
2/ When installing the python dependencies (pip install -r requirements.txt), I had a version resolution error, between diff-gaussian-rasterization 0.0.0 and diff-gaussian-rasterization 0.0.0. When looking into the submodules folder, I noticed that the package installed in the subfolder diff-gaussian-rasterization-w-pose was named diff-gaussian-rasterization (identically to the one installed in the subfolder diff-gaussian-rasterization). So I assumed that the issue came from this package name duplication. As a work-around, I changed the package name in submodules/diff-gaussian-rasterization-w-poses/setup.py to diff-gaussian-rasterization-w-pose, and renamed the subfolder submodules/diff-gaussian-rasterization-w-poses/diff-gaussian-rasterization, and the error was gone.
I should point out that I was working in a conda virtual environment.
Can you reproduce this error, or did I do something wrong ? If you can reproduce this error, is this work-around correct, and could you make the necessary modifications ?
3/ Lastly, I was working with torch 2.6.0 cu11.8, and in this last version, there was a behavior modification for the function torch.load(): specifying the argument weights_only=False is now mandatory to load arbitrary data from saved torch models (it was "only" deprecated in earlier torch versions, and now causes an error). This argument is not specified in your code, and caused an error when I tried to run your code. As a work-around, I manually added weights_only=False at the code locations leave_one_out_stage2.py:35, leave_one_out_stage2.py:129, and in cldm/model.py:18.
Can you reproduce this error ? If so, could you add these arguments, and check if it works with the last torch versions ?"
Please don't hesitate to contact me if you need further information or if anything is unclear. Thank you for your time !
Gaspard Thevenon
Hello,
I tried to contact you by mail for your Replicability Stamp submission, but didn't get a response. Just in case, I'll copy-paste the issues I had below.
"1/ When downloading the required models using your script (
models/download_hf_models.py), I had an error for the first modelv1-5-pruned.ckpt:As a work-around, I downloaded this file manually from this link (it looked like the same model):
https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5/blob/main/v1-5-pruned.ckpt
Do you know if this permission issue is an error on my side ? If not, could you either change the script, or add instructions on how to download this file manually ?
2/ When installing the python dependencies (
pip install -r requirements.txt), I had a version resolution error, betweendiff-gaussian-rasterization 0.0.0anddiff-gaussian-rasterization 0.0.0. When looking into the submodules folder, I noticed that the package installed in the subfolderdiff-gaussian-rasterization-w-posewas nameddiff-gaussian-rasterization(identically to the one installed in the subfolderdiff-gaussian-rasterization). So I assumed that the issue came from this package name duplication. As a work-around, I changed the package name insubmodules/diff-gaussian-rasterization-w-poses/setup.pytodiff-gaussian-rasterization-w-pose, and renamed the subfoldersubmodules/diff-gaussian-rasterization-w-poses/diff-gaussian-rasterization, and the error was gone.I should point out that I was working in a conda virtual environment.
Can you reproduce this error, or did I do something wrong ? If you can reproduce this error, is this work-around correct, and could you make the necessary modifications ?
3/ Lastly, I was working with
torch 2.6.0 cu11.8, and in this last version, there was a behavior modification for the functiontorch.load(): specifying the argumentweights_only=Falseis now mandatory to load arbitrary data from saved torch models (it was "only" deprecated in earlier torch versions, and now causes an error). This argument is not specified in your code, and caused an error when I tried to run your code. As a work-around, I manually addedweights_only=Falseat the code locationsleave_one_out_stage2.py:35,leave_one_out_stage2.py:129, and incldm/model.py:18.Can you reproduce this error ? If so, could you add these arguments, and check if it works with the last torch versions ?"
Please don't hesitate to contact me if you need further information or if anything is unclear. Thank you for your time !
Gaspard Thevenon