Re: Packages that work with LLMs - OLMo seems problematic

On 8/14/26 7:32 AM, Jean Louis wrote:
> I will verify more of the LLMs which claim to be trully free, such as:
> 
> - Moxin github.com/moxin-org/Moxin-LLM -- but trained with the non-free 
> LLM?
> - OLMo (from Allen AI) allenai.org/olmo
> - Marin github.com/dlwh/marin
> - DaVinci github.com/GAIR-NLP/daVinci-LLM
> - SmolLM github.com/huggingface/smollm (good candidate)
> - OpenCoder opencoder-llm.github.io
> 
> ...
> 
> Jean Louis

I just looked into Olmo, seems problematic to me. This is their paper:

https://arxiv.org/html/2512.13961v2 "This includes Dolma 3, our 
pretraining data mix encompassing carefully-sampled natural data from 
crawled sources, our midtraining mix of high-quality data designed to 
jump-start reasoning, and a large collection of science-focused PDF 
documents that unlock long-context support in Olmo"

Dolma 3 seems to be situated here:

https://github.com/allenai/dolma3

Dolma 3 appears to contain, among other things, CraneCode: 
https://github.com/allenai/dolma3/tree/main/datasets/dolma3_dolmino_mix/cranecode 
And this is listed to include as source materials e.g. "The Stack".

The stack is linked from CraneCode, and apparently found here:

https://huggingface.co/datasets/bigcode/the-stack-v2-train-smol-ids

Quote: "The Stack v2 is a collection of source code from repositories 
with various licenses. Any use of all or part of the code gathered in 
The Stack v2 must abide by the terms of the original licenses, including 
attribution clauses when relevant. "

This doesn't sound to me like attribution-free licenses at all.

Therefore, I don't understand how Olmo could be possibly free, when it 
seems to have the same problem of potentially ignoring attribution 
requirements of the training sources and therefore (at least in spirit) 
break FOSS licenses.

However I'm not a lawyer, best if you check the links for yourself to 
draw your own conclusion.

Regards,

Ellie