TANGO icon indicating copy to clipboard operation
TANGO copied to clipboard

No cached_graph fetched from huggingface_hub snapshot_download

Open t-sekai opened this issue 8 months ago • 3 comments

Hi TANGO's authors,

I'm fascinated by your work in this paper, so I decided to try and run it on my computer. First thing, I found out that the huggingface spaces was down saying "internal server error". Nevertheless, I ran the setup in the readme and reached to the step when I run app.py. When I ran the app.py program, I was able to get the gradio working. However, once I try to run the example audio and video, I got IOError like in #10 . Upon inspection, I found that the cached_graph was not in the datasets folder. Turned out, it was never downloaded using download.utils.py. When I tried downloading that snapshot individually, I fetched nothing from it. I suspect if the resources is missing on huggingface. Please help!

t-sekai avatar Apr 24 '25 06:04 t-sekai

For those who made this work, can someone shed light on which specific step is supposed to create or download the file speaker1.pkl?

File "/root/TANGO/app.py", line 179, in test_fn graph = igraph.Graph.Read_Pickle(fname=pool_path) File "/root/env3920/lib/python3.9/site-packages/igraph/io/files.py", line 223, in _construct_graph_from_pickle_file raise IOError( OSError: Cannot load file. If fname is a file name, that filename may be incorrect: /root/TANGO/datasets/cached_graph/youtube_test/speaker1.pkl.

cyberbrian avatar May 09 '25 13:05 cyberbrian

I just ran into the exact same problem, where it is looking for speaker1.pkl. I backtracked through the history and saw the following. Unfortunately, restoring the "speaker1.pkl" file did not enable the program to complete. The 3 Gig file isnt found.

Image

MrEdwards007 avatar May 25 '25 23:05 MrEdwards007

For some reason, the author deliberately deleted all pkl files which carry info of specific test targets March of 2025 Therefore TANGO/datasets/cached_graph folder is all missing Without these pkls in this folder, TANGO could not work for directly testing without training.

Following commands will lead you to old version of TANGO:

git clone https://huggingface.co/spaces/H-Liu1997/TANGO.git cd TANGO git checkout f81f0f63e697d6c34a20879d2a73489d69d6b9ce git lfs pull

If all commands work, you'll get the cached_graph folder

If lfs is not available, install and init it: apt-get install git-lfs git lfs install

For China mailand users, replace huggingface by hf-mirror export HF_ENDPOINT=https://hf-mirror.com git clone https://hf-mirror.com/spaces/H-Liu1997/TANGO.git

richardchenbuaa avatar Jun 10 '25 08:06 richardchenbuaa