support for cuda 121

[larochef]

Description

I use pytorch, have cuda 121 installed on my computer and cannot easily downgrade. My computer is running archlinux x64.

How could I get djl to use the gpu?

Will this change the current api? How? it shouldn't change the current api

Who will benefit from this enhancement? everyone running the latest cuda version

References

• list reference and related literature
• list known implementations

[frankfliu]

I didn't try it myself, but you can try to set environment variable or system properties:

export PYTORCH_FLAVOR=cu117-precxx11

or

System.setProperty("PYTORCH_FLAVOR", "cu117-precxx11");

[larochef]

I gave it a try but got:

WARN ai.djl.pytorch.jni.LibUtils - No matching cuda flavor for linux-x86_64 found: cu121.
DEBUG ai.djl.pytorch.jni.LibUtils - Loading native library: ~/.djl.ai/pytorch/1.13.1-cpu-linux-x86_64/libc10.so
DEBUG ai.djl.pytorch.jni.LibUtils - Loading native library: ~/.djl.ai/pytorch/1.13.1-cpu-linux-x86_64/libgomp-52f2fd74.so.1
DEBUG ai.djl.pytorch.jni.LibUtils - Loading native library: ~/.djl.ai/pytorch/1.13.1-cpu-linux-x86_64/libtorch_cpu.so
DEBUG ai.djl.pytorch.jni.LibUtils - Loading native library: ~/.djl.ai/pytorch/1.13.1-cpu-linux-x86_64/libtorch.so
DEBUG ai.djl.pytorch.jni.LibUtils - Loading native library: ~/.djl.ai/pytorch/1.13.1-cpu-linux-x86_64/0.21.0-libdjl_torch.so

[frankfliu]

@larochef You can try it again with 0.22.0-SNAPSHOT release.

[larochef]

@frankfliu I tested it and it worked, thx