L. Grobol

Couldn't that be a function in scipy? Something like `sparse_linear_assignment`? It would probably be useful for other things.

Or it could be an option to the existing function for when we know that the bipartite graph is sparsely connected. Also what kind of speedups do you see with...

Hi, @Narsil, so currently I do this in [zeldarose](https://github.com/LoicGrobol/zeldarose): 1. We load, tokenize (truncating) and encode text data on a single node and store it as a [datasets](https://github.com/huggingface/datasets) object at...

Hi, I'm not sure I understand > Adding pad_encoding to the bindings is doable, but your current code is even better, since you don't even have to tokenize anymore when...

Oh, actually what I do here is via `transformers` tokenizers, but I figured that getting a Python binding for the Rust padding function would be a good first step :-)

@maxstoyanov I think the idea is that in many cases, the updates in the templates (things like adapting to new dependencies, improvements to base structure, new functions) and the customisations...

I suppose that ship has long sailed but there is another alternative: - Use the hash syntax for all typst variables and functions, single letter or not - Quotes serve...

@yaksher Yeah, conversely, as someone who still uses LaTeX as well as Myst and quarto, I find it really confusing that function calls and regular text look the same: when...

Thanks for the clarification, @ArthurZucker! It still seems a bit wrong to expect the model to predict `ro_RO` given ``. The comment `wrap the last non pad token (the token)`...

Note that currently this training seems to be needed to work well with `generate` in e.g. mBART, which uses ` LANGID` as its forced prompt (loss on the LANGID could...