Paul Fultz II
Paul Fultz II
What happens to the performance for sizes like `{float, 2, 32, 512}` and `{float, 2, 32, 1024}`?
> What happens to the performance for sizes like {float, 2, 32, 512} and {float, 2, 32, 1024}? Actually these sizes are too small, as they still use block size...
Since its quantizing the weights we dont need to use our quantizer. Instead we would just take the range of weights and compute scale so it fits in the range...
> case when scales and zero point tensors of same dimension as the input tensor: I dont think the scales and zero points are ever the same shape as the...
> > * Blocked quantization: The scales and zero points have the same rank but different dimensions than the input. The `axis` of the dimension that is different than the...
In addition, we need to add unit tests for this type. `check_shapes` needs to be updated to throw an error on these types with a flag to allow it for...
Also, we need update the bytes calculation to be the correct amount. We also need to throw when calling index calculations.
> PCRE is super fast and std::regex is probably the slowest implementation in history so it is basically unusable. Yes `std::regex` is really slow, and wont be fixed since it...
> There are cases where PCRE Rules are still helpful, for example to enforce some very simple rules, such as forbidding special keyword usage like goto. Its really easy to...
Do you have a backtrace for the debug build as this does not seem to be the right place to check for the null config?