Import using:
import torchruntime
You can use the command line:
python -m torchruntime install <optional list of package names and versions>
Or you can use the library:
torchruntime.install(["torch", "torchvision<0.20"])
You can use the device database built into torchruntime
for your projects:
from torchruntime.device_db import get_gpus
gpus = get_gpus() # Returns a list of `torchruntime.device_db.GPU` instances containing the fields: vendor_id, vendor_name, device_id, device_name, is_discrete
Important: This API could break in the future, so if you're writing a program using this, please open a new Issue on this repo and let me know what you're trying to do.
This will return the recommended torch platform to use for the PC. It will analyze the GPUs and OS on the PC, and suggest the most-performant version of torch for that.
E.g. cu124
or rocm6.1
or directml
or ipex
or xpu
or cpu
.
from torchruntime.platform_detection import get_torch_platform
torch_platform = get_torch_platform(gpus) # use `torchruntime.device_db.get_gpus()` to get a list of recognized GPUs