Can't install paddle

[eddprogrammer]

I tried to install with command below.

pip install paddlepaddle-gpu==3.0.0b2 -i https://www.paddlepaddle.org.cn/packages/stable/cu123/

But kept getting error below. Looks like it is having issue download nvidia-cufft-cu12==11.2.1.3. Any workaround ?

Collecting nvidia-cudnn-cu12==9.0.0.312 (from paddlepaddle-gpu==3.0.0b2)
Downloading https://paddle-whl.bj.bcebos.com/stable/cu123/nvidia-cudnn-cu12/nvidia_cudnn_cu12-9.0.0.312-py3-none-win_amd64.whl (685.0 MB)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 685.0/685.0 MB 7.2 MB/s eta 0:00:00
Collecting nvidia-cublas-cu12==12.3.4.1 (from paddlepaddle-gpu==3.0.0b2)
Downloading https://paddle-whl.bj.bcebos.com/stable/cu123/nvidia-cublas-cu12/nvidia_cublas_cu12-12.3.4.1-py3-none-win_amd64.whl (446.2 MB)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 446.2/446.2 MB 7.4 MB/s eta 0:00:00
Collecting nvidia-cufft-cu12==11.2.1.3 (from paddlepaddle-gpu==3.0.0b2)
WARNING: Retrying (Retry(total=4, connect=None, read=None, redirect=None, status=None)) after connection broken by 'ReadTimeoutError("HTTPSConnectionPool(host='paddle-whl.bj.bcebos.com', port=443): Read timed out. (read timeout=15)")': /stable/cu123/nvidia-cufft-cu12/nvidia_cufft_cu12-11.2.1.3-py3-none-win_amd64.whl
WARNING: Retrying (Retry(total=3, connect=None, read=None, redirect=None, status=None)) after connection broken by 'ReadTimeoutError("HTTPSConnectionPool(host='paddle-whl.bj.bcebos.com', port=443): Read timed out. (read timeout=15)")': /stable/cu123/nvidia-cufft-cu12/nvidia_cufft_cu12-11.2.1.3-py3-none-win_amd64.whl
WARNING: Retrying (Retry(total=2, connect=None, read=None, redirect=None, status=None)) after connection broken by 'ReadTimeoutError("HTTPSConnectionPool(host='paddle-whl.bj.bcebos.com', port=443): Read timed out. (read timeout=15)")': /stable/cu123/nvidia-cufft-cu12/nvidia_cufft_cu12-11.2.1.3-py3-none-win_amd64.whl
WARNING: Retrying (Retry(total=1, connect=None, read=None, redirect=None, status=None)) after connection broken by 'ReadTimeoutError("HTTPSConnectionPool(host='paddle-whl.bj.bcebos.com', port=443): Read timed out. (read timeout=15)")': /stable/cu123/nvidia-cufft-cu12/nvidia_cufft_cu12-11.2.1.3-py3-none-win_amd64.whl
WARNING: Retrying (Retry(total=0, connect=None, read=None, redirect=None, status=None)) after connection broken by 'ReadTimeoutError("HTTPSConnectionPool(host='paddle-whl.bj.bcebos.com', port=443): Read timed out. (read timeout=15)")': /stable/cu123/nvidia-cufft-cu12/nvidia_cufft_cu12-11.2.1.3-py3-none-win_amd64.whl
ERROR: Could not install packages due to an OSError: HTTPSConnectionPool(host='paddle-whl.bj.bcebos.com', port=443): Max retries exceeded with url: /stable/cu123/nvidia-cufft-cu12/nvidia_cufft_cu12-11.2.1.3-py3-none-win_amd64.whl (Caused by ReadTimeoutError("HTTPSConnectionPool(host='paddle-whl.bj.bcebos.com', port=443): Read timed out. (read timeout=15)"))

[GreatV]

The error you're encountering (ReadTimeoutError when trying to download NVIDIA dependencies like nvidia-cufft-cu12) suggests that your system is struggling to download the required wheel files from the PaddlePaddle-hosted repository. This could be caused by a slow or unstable internet connection or temporary server-side issues. Below are some potential workarounds:

---

1. Increase the pip Timeout

By default, pip has a timeout for downloads set to 15 seconds. You can increase this timeout to allow for slower connections. Try the following command:

pip install paddlepaddle-gpu==3.0.0b2 -i https://www.paddlepaddle.org.cn/packages/stable/cu123/ --default-timeout=100

This increases the timeout to 100 seconds, giving more time for the large files to download.

---

2. Manually Download Missing Dependencies

If the nvidia-cufft-cu12 wheel file is failing to download, you can manually download it from the URL provided in the error message and install it locally. Here's how:

1. Visit the URL from the error message:

   https://paddle-whl.bj.bcebos.com/stable/cu123/nvidia-cufft-cu12/nvidia_cufft_cu12-11.2.1.3-py3-none-win_amd64.whl
   

   Download the file manually using a browser.

2. Install the downloaded file using pip:

   pip install path/to/nvidia_cufft_cu12-11.2.1.3-py3-none-win_amd64.whl

3. Repeat this process for any other dependencies that fail to download.

4. After installing the missing dependencies, retry the PaddlePaddle installation:

   pip install paddlepaddle-gpu==3.0.0b2 -i https://www.paddlepaddle.org.cn/packages/stable/cu123/

---

3. Switch to a Mirror or Alternate Repository

The PaddlePaddle repository you're using may be experiencing high traffic or outages. Try switching to a different mirror. For example:

pip install paddlepaddle-gpu==3.0.0b2 -i https://mirror.baidu.com/pypi/simple

Alternatively, look for an official mirror closer to your geographic region.

---

4. Use a Prebuilt Version

If manually resolving dependencies seems cumbersome, you can try using a prebuilt version of PaddlePaddle-GPU that includes all dependencies. Check the official PaddlePaddle website or documentation for standalone installers or Docker images that bundle the required dependencies.

---

5. Verify CUDA Toolkit and Compatibility

Ensure that your system has a compatible version of the CUDA Toolkit installed. For paddlepaddle-gpu==3.0.0b2 (with cu123), you need:

• CUDA 12.3
• Matching NVIDIA drivers

Run the following commands to verify compatibility:

nvcc --version

nvidia-smi

If your CUDA version or drivers are incompatible, install the correct versions from the NVIDIA website.

---

6. Fallback to Stable Version

If you are unable to resolve the issue with the beta version (3.0.0b2), consider installing a stable version of PaddlePaddle-GPU instead:

pip install paddlepaddle-gpu==2.5.0 -i https://www.paddlepaddle.org.cn/packages/stable/cu123/

---

7. Check for Firewall or VPN Issues

If you're behind a corporate firewall, proxy, or using a VPN, they could be interfering with the download. Try the following:

• Temporarily disable your VPN or proxy.
• Configure pip to use your proxy:

  pip install --proxy="http://proxy.address:port" paddlepaddle-gpu==3.0.0b2 -i https://www.paddlepaddle.org.cn/packages/stable/cu123/

---

Summary

1. Increase the pip timeout.
2. Manually download and install missing dependencies.
3. Switch to a different repository or mirror.
4. Verify your CUDA and driver compatibility.
5. Consider using a stable version if the beta version fails.
6. Check your network settings for potential interference.

If none of these solutions resolve the issue, please provide additional logs or context, and feel free to ask for further assistance!

---

Response generated by feifei-bot | chatgpt-4o-latest

[eddprogrammer]

Awesome! Increasing the pip timeout fixed it!