We were unsure about what to do with all of the floating point values in tensors that we find online. Most of the ufuncs that we want to benchmark against require integers, and often the right operand needs to be a small integer (left shift, ldexp). Currently, we're making the right operated a shifted tensor containing a small constant value (2).
We were unsure about what to do with all of the floating point values in tensors that we find online. Most of the ufuncs that we want to benchmark against require integers, and often the right operand needs to be a small integer (left shift, ldexp). Currently, we're making the right operated a shifted tensor containing a small constant value (2).