a weird problem "class 'urllib.error.URLError'"

[ning-2022]

Hi Model-angelo experts,

When I used the model_angelo to predict the protein sequence with a MRC map only, some error had appeared as following that "File "/data/wangning/anaconda3/envs/model_angelo/lib/python3.9/urllib/request.py", line 494, in _call_chain result = func(*args) │ └ (<urllib.request.Request object at 0x7fb18f8d4730>,) └ <bound method HTTPSHandler.https_open of <urllib.request.HTTPSHandler object at 0x7fb18f8bb430>> File "/data/wangning/anaconda3/envs/model_angelo/lib/python3.9/urllib/request.py", line 1389, in https_open return self.do_open(http.client.HTTPSConnection, req, │ │ │ │ │ └ <urllib.request.Request object at 0x7fb18f8d4730> │ │ │ │ └ <class 'http.client.HTTPSConnection'> │ │ │ └ <module 'http.client' from '/data/wangning/anaconda3/envs/model_angelo/lib/python3.9/http/client.py'> │ │ └ <module 'http' from '/data/wangning/anaconda3/envs/model_angelo/lib/python3.9/http/init.py'> │ └ <function AbstractHTTPHandler.do_open at 0x7fb2375c1d30> └ <urllib.request.HTTPSHandler object at 0x7fb18f8bb430> File "/data/wangning/anaconda3/envs/model_angelo/lib/python3.9/urllib/request.py", line 1349, in do_open raise URLError(err) └ <class 'urllib.error.URLError'>

urllib.error.URLError: <urlopen error [Errno 111] Connection refused> " Does It seem that model_angelo try to connect the internet? I don't understand this error. I used GPU 2080Ti and CUDA11.5 for this. Please help me. Thanks in advance!

Best regards, Ning

[jamaliki]

Hi, If this is the first time you are trying to use either the build or build_no_seq commands, then it needs to download the model weights. Are you connected to the internet?

[ning-2022]

Hi Jamaliki,

 Nice to hear from you.  Yes, your answer is very correct. Actually, this machine installed with model_angelo has an unstable connection with the internet. I will try another way.  Thanks so much for your answer. It really helps me.  

All best, Ning

[ning-2022]

Hi Jamaliki,

 Sorry to bother you again. I have tried to change the internet environment for my machine. But it failed to download the model weights. Could I download  this model weights by myself via git or wget commands and equip them into somewhere? Please give me some suggestions.  Thanks!

All best, Ning

[jamaliki]

Hi,

Yes you can download them by yourself. Although I am not sure which part is giving an error, is the ESM model downloading or the ModelAngelo weights. So, could you upload your model_angelo.log file?

[jamaliki]

In the meantime, here is a bash script to download the weights yourself (courtesy of GPT-4 ;) ):

#!/bin/bash

# Check if the bundle_name is provided
if [ -z "$1" ]; then
    echo "Usage: $0 <bundle_name>"
    echo "Available bundle names: original, original_no_seq, small_gpu"
    exit 1
fi

# Define download links
declare -A bundle_name_to_link
bundle_name_to_link=(
    ["original"]="https://zenodo.org/record/7733060/files/original.zip"
    ["original_no_seq"]="https://zenodo.org/record/7733060/files/original_no_seq.zip"
    ["small_gpu"]="https://zenodo.org/record/7733060/files/small_gpu.zip"
)

# Set the bundle_name
bundle_name="$1"

# Determine torch.hub directory
if [ -n "$TORCH_HOME" ]; then
    torch_hub_dir="$TORCH_HOME/hub"
else
    torch_hub_dir="${XDG_CACHE_HOME:-$HOME/.cache}/torch/hub"
fi

# Create destination directory
dest="$torch_hub_dir/checkpoints/model_angelo/$bundle_name"
mkdir -p "$(dirname "$dest")"

# Check if the file already exists
if [ -f "$dest/success.txt" ]; then
    echo "Bundle $bundle_name already exists."
    exit 0
fi

# Download and extract the zip file
echo "Setting up bundle with name: $bundle_name for the first time."
wget -q -O "${dest}.zip" "${bundle_name_to_link[$bundle_name]}"
unzip -q "${dest}.zip" -d "$(dirname "$dest")"
rm "${dest}.zip"

# Create success.txt file
touch "$dest/success.txt"
echo "Successfully downloaded model" > "$dest/success.txt"

echo "Bundle $bundle_name successfully installed."

[jamaliki]

You should run the script above with the model_angelo conda environment activated twice. Once for original and once for original_no_seq

[ning-2022]

Thanks for your quick replies. The script you provided is working to download the weights now. I'm not sure whether this connection error relates about ESM model or model_weights downloading. Here is my error log file. Thanks for your time.
model_angelo.log

[jamaliki]

Let me know if you have any more issues!