Environment Configuration Overview
All aifare platform images are based on Ubuntu, specifically Ubuntu 22.04.
The platform comes pre-installed with mainstream AI frameworks and multiple versioned images. Once an instance is launched, the corresponding framework environment is readily available without manual setup. If the built-in versions do not meet your needs, you can customize the environment as described below.
Pre-installed Mainstream Frameworks and Versions
| Framework | Version | CUDA Version | Python Version | Notes |
|---|---|---|---|---|
| PyTorch | 2.1.0 | 11.8 | 3.11 | torchvision/torchaudio included |
| PyTorch | 2.2.0 | 11.8 | 3.11 | |
| PyTorch | 2.3.0 | 11.8 | 3.11 | |
| PyTorch | 2.4.0 | 12.1 | 3.12 | |
| PyTorch | 2.5.0 | 12.1 | 3.12 | |
| PyTorch | 2.5.1 | 12.4 | 3.12 | |
| PyTorch | 2.7.1 | 12.6 | 3.12 | |
| TensorFlow | 2.16.1 | 12.0 | 3.11 | |
| TensorFlow | 2.17.0 | 12.0 | 3.11 | |
| TensorFlow | 2.18.0 | 12.0 | 3.11 | |
| PaddlePaddle | 2.6.2 | 12.0 | 3.11 | paddlepaddle-gpu |
Recommended Usage Guidelines
- Prefer built-in platform images: If your required version of Torch, TensorFlow, etc. is already provided, use them directly for a ready-to-use experience.
- Customize frameworks/versions: If a specific version is required (e.g., PyTorch=1.9.0 + CUDA=11.1), use the appropriate CUDA version of the Miniconda image and install the framework via
pip. - Customize Python version: If a specific Python version is needed, you can install it manually inside a Miniconda image.
- Custom CUDA installations: Refer to the platform documentation for installing other CUDA versions manually.
Related Documentation
For special environment requirements, feel free to contact aifare support or refer to the platform's community resources.