Sludge

Hmm, it's still broken with 3+ entries

There we go. The fixup should only happen when an index was removed that is *not* at the end of the array.

There are no `Gemm` or `MatMul` nodes in the network. In fact let me post the `nnx info` output for reference: ``` +------------------+----------------------------------------------------+ | Model version | 0 | +------------------+----------------------------------------------------+...

No, it doesn't work yet. Support for grouped convolutions is required to make this model work properly. I tried to implement them a few weeks ago but couldn't get them...

One long-term feature I'd like to see is having wonnx expose its input `wgpu::Buffer` so that the application can use the GPU to produce the inference input instead of always...

Thanks for the quick feedback!

This was fixed by https://github.com/webonnx/wonnx/pull/157

*Any* type alias of a `struct` will reproduce it. ```rust pub type Alias = Vec; ``` Result: 

To me, "aliased type" strictly refers to the right-hand side of the `type A = B` syntax, so I would find this feature much less confusing if it simply used...

Ah, I looked into https://github.com/tokio-rs/tracing/blob/master/tracing-log/CHANGELOG.md, which hasn't been updated