tract
tract copied to clipboard
sonos
Tiny, no-nonsense, self-contained, Tensorflow and ONNX inference
Re: #787, poke @kali This is a dump of my AVX512f branch with kernels and some notes. I'll be cleaning this up a bit more before I think it's fine...
So I have this einsum operator that broadcasts the second parameter (self.weight) to be able to compute the expression. ```py x = torch.einsum("btgi,gih->btgh", x, self.weight) ``` This results in the...
Hi, I've been trying to optimise my model, or more specifically optimise usage of it in my Rust code. And I thought that using batch should improve the performance, but...
Hey @tgolsson! The AVX512F question is back on the radar with another team evaluating performance on this architecture. IIRC, you have already done the heavy lifting on this front, with...
Hello :) This PR is made to make the yolov5 postprocessing (which uses the If operator with a non-constant condition!). This PR is not supposed to be merged as-is, as...
Maybe I was a bit too quick to close #718 as it seems to still have some issues when running depending on exact flags I pass. I'll post these in...
Add the ability to specify a quantized datum (`QI8`, `QU8`, `QI32`) type when specifying input facts. It can be specified with the following pattern: - QI8 with zp scale (zp:...
