Can we run lama3-8b on other npu ?
Can we run llama3-8b on other Npu on devices such as Qualcomm's Snapdragon X or Intel's Core Ultra?
can we have one custom llama model which runs across all 3 npu?
Hello
At least for this model, it is assumed that it will run using AMD libraries, so I don't think it will work.
It may be possible to make it work by converting it, but since the formats supported by each library are different, I think it is currently impossible to unify them.
However, it is better to check the details officially.
https://huggingface.co/amd
https://huggingface.co/qualcomm
https://huggingface.co/Intel
ya, that's the issue is formats are supported by different libraries are different.
is it possible to convert the llama to onnx then can we run it across different libraries?
Conversion is possible, but when I tried it before, the performance deteriorated, so I felt that the ideal of onnx was not realized.