Daniel D. Johnson
Daniel D. Johnson
Perhaps one could daisy-chain that with something like [tm2deftheme](https://github.com/emacsfodder/tmtheme-to-emacs) or [coloration](https://github.com/sickill/coloration)?
An interesting property of inferred superclass instances: it allows refining the class hierarchy without breaking user code. In particular you could lift operators from an interface into a new parent...
I agree that this isn't as well documented as it could be; I was mostly focused on trying to get something that worked. I'll take another pass through it and...
I believe you've linked to the prototype of pure-Dex userspace parsing, which isn't actually used when parsing normal Dex code. But indeed the Haskell parser implementation that parses Dex code...
Yeah, the current syntax doesn't support record flattening. I think it's actually a more fundamental issue than that, too, because the concatenation of two records can't be unified with a...
Hm, interesting idea! One question is what the API for this should be. Some ideas: - **Add a keyword argument to `nmap`:** Something like `nmap(fn, batch_sizes={"foo":2, "bar":4})`. This would be...
I agree that having a single `scan` at the outside seems better than having a number of smaller scans in the inside for this use case. Thanks for running the...
Hm, I'm not sure this is necessary? You can provide the device and sharding for the array before wrapping it, and `pz.nx.wrap` should preserve it, e.g. ```python arr = pz.nx.wrap(jax.device_put(my_array,...
Agreed this would be a very useful feature! I think it should be pretty easy to prototype something like this without needing to directly change Penzai's implementation, because Penzai is...
Sorry for the confusion! This seems like an artifact of how Python imports work. Right now importing `penzai.nn` does NOT recursively import all submodules of `penzai.nn`. However, once somebody imports...