Mathieu Poumeyrol
Mathieu Poumeyrol
Found them. So I can at least right now clarify what happens with text.onnx. Now we are (somewhat) correctly reading symbols from models. The error you get is a conflict...
Ok, this panic is just due to the fact that we are trying to access a tensor by index with a value bigger than the axis dimension. This is pretty...
The more the merrier... new command line argument: --random-range `cargo run -p tract -- text.onnx --set b=1 -O run --const --allow-random-input --random-range 1.0..1000.0` So this works for me. Do we...
I'm curious, what does yolo do that needs a non-constant if conditional ?
No worries here. I fully agree that "common" networks should run out of the box in tract. Within reason of course: Tensor List and/or Sequence support would be another story...
Hello, thanks for your interest in tract. It always depends on the model, but indeed, batch should rarely slower than simple iteration. For image models, the gain could be marginal,...
We don't have pulsification for EinSum yet. It is not necessarily super-hard to do, so I can have a look, but it would help if you could make a test...
Hey @tgolsson, thanks for making this available. I fixed a couple of correctness issues here and there, and put on a simple heuristic to pick a avx512f mmv. But I...
> I might've forgotten a stash or two, but I did not have a complete set of kernels. I found the workflow for the FMA kernels quite frustrating when I...