First time installing any AI model and I’m basically just following a simple guide to get stable diffusion 1.5. However, I noticed once command prompt got to the point where it needs to download this model from huggingface, my download speeds drop from the usual 5 mB/s down to like 200 kB/s.
I did some testing and went on the website itself trying to download the models through my browers and even then, I’m still hitting speeds of 200 kB/s. I noticed that basically every model I try to download on the most liked page was getting these speeds. However, I then tried picking some random models, on less liked pages and my speed shot up, back to my full speed of 5mB/s. Is hugging face throttling downloads on their most popular model/potentially unable to handle the download traffic for their most popular models?
Just a little concerned about having to wait 6 hours to download any of the well known models. My internet could cut out at any time and maybe I’ll lose all the progress.
I am getting this too, but its not limited to popular models, my speeds start around 10 mB/s then drop down to 500 kB/s. Upon cancelling and restarting the download i get a speed burst again.
I’m not sure if it’s my line, an upstream line, or HF’s server, but my connection speed has been dropping at random times today.
I’ve tried a few times and it usually works…
Did others also notice a significant slow-down in download speeds? I’m getting around 10MB/s, which really messes up my workflow (where I’m downloading into virtual machines).
there’s a neat trick to getting some substancial boost in download speed using hf-transfer
try this :
pip install hf-transfer
# setup backend to use hf_transfer (one time only)
import os
os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"
# put python script for downloading things below
from huggingface_hub import HfApi, logging
logging.set_verbosity_debug()
hf = HfApi()
hf.upload_file(path_or_fileobj="/path/to/my/repo/some_file", path_in_repo="some_file", repo_id="my/repo", repo_type="model")
you will sacrifice your progress bar but it’s based on rust so it’s faster than python.
