Re: Packages that work with LLMs - SmolLM seems problematic too

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've now looked into the SmolLM model which seems problematic too. The 
main page is listed here: SmolLM github.com/huggingface/smollm

They have an infographic, sadly it's an image only, that lists the 
following data source for SmalLM: "FineMath" This seems to be referring 
to this:

https://huggingface.co/datasets/HuggingFaceTB/finemath

Quote: "FineMath consists of 34B tokens (FineMath-3+) and 54B tokens 
(FineMath-3+ with InfiMM-WebMath-3+) of mathematical educational content 
filtered from CommonCrawl. To curate this dataset, we trained a 
mathematical content classifier using annotations generated by 
LLama-3.1-70B-Instruct. We used the classifier to retain only the most 
educational mathematics content, focusing on clear explanations and 
step-by-step problem solving rather than advanced academic papers."

As I pointed out in a previous email, Common Crawl just seems to be 
public websites with all sorts of licensing, including potentially 
proprietary. The info given has no indication that this was filtered for 
FOSS licensing, let alone licensing compatible with not giving any 
attribution as models tend not to do during operation.

Therefore, I don't really see now SmolLM's output is open either. But I 
could be wrong, check above sources for yourself, this isn't legal 
advice, etc. etc.

Regards,

Ellie