Luca Antiga

Is it just for win?

@tshu-w @vedal thanks for your comments My concern here is breaking people’s code extensively, but I’m all for finding a solution that honors compatibility (including adding a new recommended way...

`torch.bool` is actually an instance of `torch.dtype` ``` >>> type(torch.bool) ``` so you need to check if `something.dtype == torch.bool`

@csarofeen yes thanks for flagging

The failing loudly (and not obscurely) part is super important. Let's keep this issue to keep track of this aspect. I would prioritize enumerating corner cases as much as we...

#209 for reference

@t-vi what would be a similar lookaside to start from for anyone wanting to approach this?

Looks super neat, to simplify we could also do ```python jm = thunder.jit( m, transforms=MaterializationTransform("cuda", init=ref_m.state_dict()) ) ```

aah makes sense

that's right, in autocast we don't consider `device` https://github.com/Lightning-AI/lightning-thunder/blob/main/thunder/core/transforms.py#L3788 does this have practical impacts on target models?