Difference between cuda, DML and CPU onnx models?
#1
by
PaulTheHuman
- opened
Hi, can someone explain what the difference is between the Cuda, DML and CPU onnx models?
Are they interchangeable?
Since they are all in the onnx format shouldn't they all work on all devices?
Or is it simply that some work better on certain devices?
The originally uploaded models were optimized for each execution provider. Due to precision differences and operator support, they were not interchangeable and worked best for certain devices. With the newly uploaded models, there is now one optimized ONNX model for CPU and one optimized ONNX model for GPU. The work to support one unified, optimized ONNX model for both CPU and GPU is in progress.
kvaishnavi
changed discussion status to
closed