# Fame PyTorch ROCm Kit β Installation Instructions π | |
This file explains how to manually install the Fame-built PyTorch ROCm version. | |
--- | |
## π οΈ 1. Prepare Environment | |
Activate your Python environment (recommended: venv): | |
```bash | |
source ~/rocm_env/bin/activate | |
``` | |
Install CMake if missing: | |
```bash | |
pip install cmake | |
``` | |
Ensure HIP and ROCm paths are correctly set (`ROC_ENABLE_PRE_VEGA=1`, etc). | |
--- | |
## π οΈ 2. Build PyTorch | |
Navigate to the Fame Kit folder: | |
```bash | |
cd fame-pytorch-kit/ | |
``` | |
Run the build script: | |
```bash | |
bash build_torch.sh | |
``` | |
--- | |
## π¦ 3. Install the Wheel | |
Navigate to the dist folder: | |
```bash | |
cd pytorch/dist/ | |
``` | |
Install the Wheel manually: | |
```bash | |
pip install torch-2.8.0a0+gitc402b3b-cp312-cp312-linux_x86_64.whl | |
``` | |
--- | |
## β 4. Verify Installation | |
Test your installation: | |
```bash | |
python -c "import torch; print(torch.cuda.is_available())" | |
python -c "import torch; print(torch.cuda.get_device_name(0))" | |
python -c "import torch; print(torch.version.hip)" | |
``` | |
You should see: | |
- `True` (GPU detected) | |
- Your GPU model name | |
- HIP version string | |
--- | |
# Built with Fire. Built with Fame. π‘οΈπ₯ | |