Guillaume Lagrange
Guillaume Lagrange
> @nathanielsimard > > I found a way to fix this, and now it works for `burn` package. However, it is just to re-export all things in `burn_core` and `burn_train`...
Looking at the [panicking statement](https://github.com/tracel-ai/burn/blob/v0.13.2/crates/burn-import/src/onnx/dim_inference.rs#L281) in question, seems to be an issue with the `unsqueeze` op. Right now most of the ONNX ops are parsed during import expecting some of...
In previous versions the unsqueeze onnx op was also checking the input shapes but it wasn't required. This has been changed on main if you want to try it out....
Well we also use hashmap as a drop-in replacement for no-std, so I understand their point. It's also a public facing API so that would be a breaking change, and...
Just to address the OP: - `tch` errors are strictly for `unimplemented!` deform conv2d, so this is expected (though we should mark these tests differently or not run them with...
Ahh thanks for the pointer. I tried to get access to a mac by stealing my girlfriend's old macbook but it's too old and doesn't have any of the devtools...
> I think this is [tracel-ai/cubecl#909](https://github.com/tracel-ai/cubecl/issues/909). [8ca52c9](https://github.com/tracel-ai/burn/commit/8ca52c9abdccd9eb0d449d56511f83a67f5b8132) doesn't break it per se, but before that, subgroups were accidentally NOT enabled on metal, and that commit correctly enables them on Mac....
Root: ``` thread '' panicked at /home/schneider/.cargo/registry/src/index.crates.i159f17d22bba15001f/burn-tensor-0.16.0/src/repr/handle.rs:73:32: 9996/10000 2192= == PANIC ===handle for tensor TensorId { value: 560132 } ``` Seems very much related to [this user's error posted on...
Ahh right, we've seen this issue with the update to wgpu 0.24 that has more complex types. It can be solved by increasing the default recursion limit (128) at the...
The problem is most likely in the types used for burn-import and the tests. It should be based on the backend int elem type. Would have to take a look...