Guillaume Lagrange

> I tried [this](https://github.com/openppl-public/ppl.nn/blob/master/docs/en/onnx-model-opset-convert-guide.md) and it did not help... That said from the comments my understanding is that the `Only tensor indices is valid` issue isn't due to an old...

> Used `onnx-simplifier` (https://github.com/daquexian/onnx-simplifier) which removes lots of these crazy gather reshape operation, which were due to lacking capabilities of the older ONNX OpSet versions. > > Here is updated...

Just a note that while the original issue was circumvented with the help of onnx-simplifier, it does not always appear to remove the problematic gather node with indices being a...

Can't self-assign to an opened PR, but just stating in the open that I will take over the WIP and complete the refactor.

Thanks for flagging this! I believe this is due to the current implementation of the BCE loss for `tensor.log()` which results in `-inf` for values of `0.0`. We need to...

Yeah we realized this the other day with the upgrade to wgpu 0.24.0.. see [this discord convo for reference](https://discord.com/channels/1038839012602941528/1146234443846127758/1331359573659091037). This seems to stem from new complex types in wgpu. As...

Some slight variations, but as long as the inputs during training don't have some weird values that deviate then I don't think that will be the issue. I'll reopen this...

> There is strange correlation between training accuracy (46%) and amount of zeroes in training set labels. The model has not converged as expected if you're getting NaNs during training,...

> use burn::{ > backend::{Autodiff, Wgpu}, > nn::loss::BinaryCrossEntropyLossConfig, > tensor::{Int, Tensor, TensorData}, > }; > > #[test] > fn test_binary_cross_entropy_preds_almost_correct() { > type MyBackend = Wgpu; > type MyAutodiffBackend =...

Ahhh ok I can reproduce on 0.16 ``` ---- test_binary_cross_entropy_preds_almost_correct stdout ---- thread 'test_binary_cross_entropy_preds_almost_correct' panicked at src\main.rs:23:17: Tensors are not approx eq: => Position 0: NaN != 0.000000849366756483505 | difference...