Andrew Brown

That first one only does a `load` and we need to walk through the entire API. Would you like to take a stab at adapting the second one to WIT?

> i agree it's better to migrate the latter one to wit. my understanding is it's mainly about migrating wasi-nn, which is https://github.com/bytecodealliance/wasi-nn/tree/main/rust. is it right? Well, there's work to...

> I am proposing we handle it the same way most of the ML frameworks handle it. More specifically, most frameworks allow passing a linear array that is re-interpreted based...

I think I have a new suggestion to add to this thread: why don't we use `tensor-dimensions` to describe the length of the `bytes` arrays laid out end-to-end in the...

@geekbeast's use case was the original motivation for this issue; since he feels like he can make do without a new type, let's park this issue until someone absolutely needs...

Like we discussed (and as commented in #70), this change is probably not desirable if we adopt https://github.com/WebAssembly/wasi-nn/pull/74: if a user can't change where a model resides any more, they...

Your feedback would be great! The primary ways to discuss are GitHub issues like this one but @mingqiusun and I are also usually online in the [BytecodeAlliance Zulip channel](https://bytecodealliance.zulipchat.com). >...

> `set_input_by_name` ... `set_output_by_name` It is unfortunate that the TF API does not match the wasi-nn API here, but there is a workaround. @brianjjones, in his PR to add TF...

I do not think (yet) that this specification should pick a single encoding format for models. ONNX may be a good choice for some users but not for others; it...

cc: @sunfishcode, @mingqiusun, @leecam, @rasquill, @jlb6740