As many know, due to network restrictions, users in mainland China cannot directly access the world's largest AI model library, https://huggingface.co. Many AI software (e.g., pyVideoTrans) require online model downloads from this source. To address this, the domestic mirror site https://hf-mirror.com is used by default.
Here's the challenge:
Recently, this mirror site has become increasingly unstable, frequently experiencing download failures and connection timeouts. It has even introduced rate limiting, often returning 429 "Too Many Requests" errors.
These issues typically manifest as the following error messages:
Don't worry! Below, I'll guide you through several simple and reliable methods to successfully download your models. Choose the method that works best for you and get it done easily.
Method 1: Download Directly from Hugging Face Official Site (Recommended for VPN/Proxy Users)
This is the most direct and stable method, provided you have access to a VPN or proxy service and can open https://huggingface.co in your browser.
Steps:
Set up Proxy: In the software's "Network Proxy" text box, enter your proxy address and port (format:
proxy_address:port
). Alternatively, configure a system-wide proxy to ensure internet access for your entire computer.Create Lock File: In the same directory as
sp.exe
, create an empty file and name ithuggingface.lock
. Important: The file extension MUST be.lock
, not.txt
! To avoid errors, please enable "File name extensions" in File Explorer (Top of folder -> View -> Check "File name extensions"), as shown below.Tip: Can't create a
.lock
file? No problem! Right-click to create a newhuggingface.txt
file, then rename it, changing.txt
to.lock
. Remember to enable "File name extensions" first, as shown above.
With this setup, the software will prioritize downloading models directly from the official site, offering faster speeds and higher success rates!
Method 2: Use Dedicated Download Tools (Super Simple, One-Click Solution)
To make downloads even easier, I've prepared two dedicated EXE tools: one for downloading from the mirror site hf-mirror.com
, and another for downloading from the official site huggingface.co
. You can choose based on your network conditions.
Steps:
Visit the download page, download the file you need, and place it in the same directory as
sp.exe
.Double-click to run the tool, and the model will automatically start downloading and be placed in the correct location.
Method 3: Manual Download from GitHub (Compressed Archives)
If the above methods don't work, you can manually download pre-packaged model archives from GitHub.
Steps:
Open in your browser: https://github.com/jianchang512/stt/releases/tag/0.0
Download the model files you need.
Extract the files to the software's
models
folder. Crucial: When extracting, double-check the directory structure to ensure it's correct!