changqi1 / xDNN

Optimized DNN kernels on X86 architecture.
Apache License 2.0
2 stars 1 forks source link

ubuntu 20.04 #1

Open LittleNoob2333 opened 1 month ago

LittleNoob2333 commented 1 month ago

sir, can I use it on ubuntu 20.04?

changqi1 commented 1 month ago

Yes, I think.

LittleNoob2333 commented 1 month ago

Yes, I think.

sir What is the relationship between xDNN and oneDNN?

changqi1 commented 1 month ago

Yes, I think.

sir What is the relationship between xDNN and oneDNN?

oneDNN is Intel's general Deep learning accelerate library. xDNN is only for LLM operators.

LittleNoob2333 commented 1 month ago

Yes, I think.

sir What is the relationship between xDNN and oneDNN?

oneDNN is Intel's general Deep learning accelerate library. xDNN is only for LLM operators.

That is to say, xDNN may perform better on LLM than oneDNN?

changqi1 commented 1 month ago

Yes, I think.

sir What is the relationship between xDNN and oneDNN?

oneDNN is Intel's general Deep learning accelerate library. xDNN is only for LLM operators.

That is to say, xDNN may perform better on LLM than oneDNN?

xDNN could do more operation than oneDNN, like input data is FP16 data type and weight data is INT4 data type. xDNN could accept it, but oneDNN need to optimize.

LittleNoob2333 commented 1 month ago

Yes, I think.

sir What is the relationship between xDNN and oneDNN?

oneDNN is Intel's general Deep learning accelerate library. xDNN is only for LLM operators.

That is to say, xDNN may perform better on LLM than oneDNN?

xDNN could do more operation than oneDNN, like input data is FP16 data type and weight data is INT4 data type. xDNN could accept it, but oneDNN need to optimize.

So, what are the differences between xDNN and oneDNN on kernels implemented with AMX, or does xDNN have specific optimizations for LLM kernels?