peterbell10

Oh I see the issue, `get_active_torch_device` has been added recently on the main branch and isn't in a released version but the docs on the website were updated to match...

The docs being for an unreleased version is a real issue though.

@pytorchbot merge

It isn't fixed yet, but will be fixed by llvm/llvm-project#113928

cc @lezcano for the layout question. +1 for `CoalescedLayout`. Indeed that seems more precise here, and would leave more space for future developments.

sem does not make the atomics more deterministic, changing rtol/atol is the only thing that makes sense.

Can you add a lit test that shows an improvement with this PR?

I've had a read through some of the docs and tried using the library for a few bindings and I've got some answers but still lots of questions. 1. It...

The prefabs specifically rely on this coercion, if you rewrite the `cLuaState::GetStackValue` for `AString` then they fail to load completely. Whether this was intentional or not though...

This was where I was seeing errors: https://github.com/cuberite/cuberite/blob/57690b81a24a29d70cb6f4196a6e0f521a3cb61b/src/Generating/PrefabPiecePool.cpp#L497-L498 "Direction" in the cubeset files are numbers which get coerced into strings when getting them from the stack by calling `lua_tostring`.