Huggingface demo problem

[ratmother]

Hi there! I'm trying to run the Huggingface demo in a duplicated space but running into an issue 'Something went wrong Connection errored out' - from the logs I don't see any errors, when I check the container log all I see is (Additionally maybe this is just a spaces thing but it seems that it re-downloads a bunch of models every time I run the demo possibly because the connection erroring out is causing some kind of reset? );

...
Collapsing Lin Lora in out_proj
Reshaping encoder.mid.attn_1.q.weight for SD format
Reshaping encoder.mid.attn_1.k.weight for SD format
Reshaping encoder.mid.attn_1.v.weight for SD format
Reshaping encoder.mid.attn_1.proj_out.weight for SD format
Reshaping decoder.mid.attn_1.q.weight for SD format
Reshaping decoder.mid.attn_1.k.weight for SD format
Reshaping decoder.mid.attn_1.v.weight for SD format
Reshaping decoder.mid.attn_1.proj_out.weight for SD format
(At this point Huggingface's log clears)
Looking in links: https://dl.fbaipublicfiles.com/pytorch3d/packaging/wheels/py38_cu117_pyt1131/download.html
Collecting pytorch3d
  Downloading https://dl.fbaipublicfiles.com/pytorch3d/packaging/wheels/py38_cu117_pyt1131/pytorch3d-0.7.3-cp38-cp38-linux_x86_64.whl (72.7 MB)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 72.7/72.7 MB 131.5 MB/s eta 0:00:00
Requirement already satisfied: iopath in /home/user/.pyenv/versions/3.8.9/lib/python3.8/site-packages (from pytorch3d) (0.1.10)
Requirement already satisfied: fvcore in /home/user/.pyenv/versions/3.8.9/lib/python3.8/site-packages (from pytorch3d) (0.1.5.post20221221)
Requirement already satisfied: Pillow in /home/user/.pyenv/versions/3.8.9/lib/python3.8/site-packages (from fvcore->pytorch3d) (9.0.1)
Requirement already satisfied: pyyaml>=5.1 in /home/user/.pyenv/versions/3.8.9/lib/python3.8/site-packages (from fvcore->pytorch3d) (6.0)
Requirement already satisfied: tqdm in /home/user/.pyenv/versions/3.8.9/lib/python3.8/site-packages (from fvcore->pytorch3d) (4.64.1)
Requirement already satisfied: tabulate in /home/user/.pyenv/versions/3.8.9/lib/python3.8/site-packages (from fvcore->pytorch3d) (0.9.0)
Requirement already satisfied: yacs>=0.1.6 in /home/user/.pyenv/versions/3.8.9/lib/python3.8/site-packages (from fvcore->pytorch3d) (0.1.8)
Requirement already satisfied: termcolor>=1.1 in /home/user/.pyenv/versions/3.8.9/lib/python3.8/site-packages (from fvcore->pytorch3d) (2.3.0)
Requirement already satisfied: numpy in /home/user/.pyenv/versions/3.8.9/lib/python3.8/site-packages (from fvcore->pytorch3d) (1.22.4)
Requirement already satisfied: portalocker in /home/user/.pyenv/versions/3.8.9/lib/python3.8/site-packages (from iopath->pytorch3d) (2.7.0)
Requirement already satisfied: typing-extensions in /home/user/.pyenv/versions/3.8.9/lib/python3.8/site-packages (from iopath->pytorch3d) (4.5.0)
Installing collected packages: pytorch3d
Successfully installed pytorch3d-0.7.3
./weights/3DFuse_sparse_depth_injector.ckpt already exists.
/home/user/.pyenv/versions/3.8.9/lib/python3.8/site-packages/gradio/deprecation.py:43: UserWarning: You have unused kwarg parameters in Row, please remove them: {'scale': 1.0}
  warnings.warn(
Running on local URL:  http://0.0.0.0:7860

To create a public link, set `share=True` in `launch()`.

[ammarhusain]

hello, I am running into the same problem!

[j0seo]

Hi, sorry for the late reply. Have you tried running it on a paid GPU environment instead of a CPU environment? Due to pytorch3d installation issues, it may only be possible to install it on a GPU environment. We confirmed that the dependency installation worked well in the A10G environment. If there's any other reason it's not working, please let me know!

[ratmother]

So I think its to do with the memory running out at an educated guess. The Nvidia A10G large works fine (on a slightly modified version of the code) but the other GPUs fail, the Nvidia A10G large has 46 GB RAM and 24GB VRAM and runs but the Nvidia A10G small 4 vCPU with 15 GB RAM 24GB VRAM does not, it could be anything though I guess (no errors makes it hard to debug). If this is intended it might be worth letting people know as the difference in price is significant.

[j0seo]

Thank @ratmother for letting me know. We didn't know this as we only tested it on A10G large in the HuggingFace environment. We will update the HuggingFace demo to let people know this.