On the parametric search space

apache / tvm

Open deep learning compiler stack for cpu, gpu and specialized accelerators

https://tvm.apache.org/

Apache License 2.0

11.6k stars 3.44k forks source link

On the parametric search space #8752

Closed jqzhao7 closed 3 years ago

jqzhao7 commented 3 years ago

Dear teacher, I am a graduate student studying TVM. First of all, As a very meaningful job, tvm, I recently encountered a problem: What is the size of the search space that TVM generates for the model waiting to be optimized? Where should I look up information on this subject? Or what is the definition of the size of the search space?

vinx13 commented 3 years ago

Relay model is lowered to a few operators in topi. Each (tunable) operator has one or more templates in topi, for example https://github.com/apache/tvm/blob/596ff5cff576d8ded1251bb936eb7198551cc481/python/tvm/topi/cuda/conv2d_direct.py#L25. You are encouraged to open a new thread on https://discuss.tvm.apache.org/ as we use the forum for related discussions.