Evaluate kernl/ triton for an ONNX alternative #15
Description
- There'as lots of promising farmeworks on the horizon that'll allow us to stick to close to Pytorch - https://github.com/ELS-RD/kernl, and https://github.com/triton-lang/triton
- Kernl are promising lower latency than onnxruntime with some setups, it'd be great to evaluate of this is the case with some of these foundation models!