Guillaume Lagrange

@younes-io did you figure out the issue? What device are you trying to use with wgpu? Perhaps a new issue could be opened to detail your problem, but I agree...

By simple CUDA program, you mean like adapting the simple example below to use your NVIDIA gpu with the wgpu backend? ```rust use burn::tensor::Tensor; use burn::backend::Wgpu; // Type alias for...

Hmm CI failed but the failing test is unrelated to this PR /edit: looking for feedback on the benchmark implementation but let's wait until we optimize some stuff before we...

@NicoZweifel Hey 👋 I'm going through opened issues/PRs right now, looks like there hasn't been a lot of activity here for a while. I'll close the draft PR but if...

Not all ops are supported for ONNX import yet, but in this case it seems to have failed at the `Concat` node (as pointed out by the panic at [this...

Right now the docs for tensors basically only mention this: > With Burn, you can focus more on _what_ the model should do, rather than on _how_ to do it....

I think we should start by tackling 1, 2 and 3. > 4. Consider automatically performing unsqueeze in certain cases to improve API usability, while being mindful of potential unintended...

> @laggui I think torch has the same behavior; it's batch matrix multiplication where the batch dimension can be broadcasted. You're right! At least, for element-wise ops but not for...

Hmmm I think I would agree that the schedule should start with the initial LR. Right now we're essentially skipping the initial step. We could possibly add a `scheduler.lr()` function...

I don't necessarily mind breaking the current usage, as long as it is valid. The suggested solution to update the state but return the previous state in `.step()` doesn't feel...