Guillaume Lagrange

Thanks for flagging this issue, you're absolutely right! The metric calculation at each batch is correct, but the aggregation is inaccurate. There are actually two problems with the way it's...

For the metric state, this can be kept in a new state type so we can accumulate the predictions and targets. But you're right that for logging an entry, this...

Just gonna give a quick summary for the setup: - The repo is set up as a workspace, so the main `Cargo.toml` is defined at the workspace level. Same for...

> Is it just > > cd examples > cargo new --lib > ? > Pretty much! You just need to make sure that the workspace readme is not inherited...

Yeah I think it makes sense to briefly describe this process, and the best place for that is absolutely in the contributor book. This will make it easier for everyone.

> Mainly bool elem type was not supported back then. I am not sure if we still have this blocker. Bool is implemented and the backend also has an associative...

The difference you're seeing here is not actually from the allocation, but overhead from the optimizations applied by autotune and fusion. It doesn't really make sense for your minimal use...

Hey 👋 glad to see you're enjoying Burn! We've had an outstanding issue (#1393) opened ever since the interpolate module ops were added to extend support. Right now only the...

Thanks for playing around with the demo 😄 It's a fairly simple model (part of our examples) that we recently improved to be more robust to some transformations on the...

> $env:TORCH_CUDA_VERSION="cu129" We are currently on tch 0.19, which requires libtorch v2.6.0 and does not support `TORCH_CUDA_VERSION="cu129"`. How are you building/linking tch with libtorch?