Re: Packages that work with LLMs

On 8/14/26 07:28, Jean Louis wrote:
 > Apertus relies on "openly available data." Note how they didn’t actually
 > define "openly." It’s just a vibe.
 >
 > Their logic is delightfully simple: if a website doesn’t explicitly tell
 > a bot to go away in its robots.txt, then the data is "openly available."

To clarify, they actually specify their data collection procedure in: 
https://arxiv.org/pdf/2509.14233

So most of it is FineWeb-2, which is hosted on Hugging Face: 
https://huggingface.co/datasets/HuggingFaceFW/fineweb-2 The dataset is 
licensed under ODC-By 1.0 which allows you to "copy, distribute, and 
modify" it.

Yet the dataset contains text that were clearly not permissevily 
licensed by its original authors.

If one thinks in terms of programs, then the dataset is part of the 
soruce code of the Apertus LLM, the weights its machine code, and 
training is compiling. You cannot reconstruct the weights without the 
dataset.


--
mvh Björn