3 ms·
If the model is closed and served through an API, probably not. But at the moment Mistral seems to focus on smaller and efficient models that you can fine-tune
by lelag 3y ago
If the model is closed and served through an API, probably not.
But at the moment Mistral seems to focus on smaller and efficient models that you can fine-tuned and run on your own infrastructure.
Mistral 7b was free but if they come up with larger capable models, they can offer them with a non-commercial free license and a sell a commercial license for companies that need to fine-tune it.
Looking at the performance of small fine-tuned models on specific tasks, I think most company will prefer taking that route rather than trusting OpenAI with their own data, the market is huge I think.
- jgalt212 3y ago> trusting OpenAI with their own data often that's a GDPR violation. there is not such thing as perfect data scrubbing.
- IanCal 3y agoYou can definitely transfer data to another processor without breaking GDPR rules, even US ones. You need to take care, but it doesn't require perfect data scrubbing.
- mark_l_watson 3y agoMy hope, being fair to Mistral, is that they keep supporting the 7B model series, and release a set of products such as: one off non-commercial and commercial model purchases of more capable 13B, 20B, and 30B models. I would also like to see a subscription service so, if for example, I was using 13B commercially, I could pay a yearly fee for updates and support for that model. EDIT: since I am retired and just obsess about LLMs and in general deep learning as a ‘gentleman scientist’ I hope Mistral has inexpensive non-commercial options for larger, more capable, and commercially supported models.
- mg 3y agoHmm... I would think the on-premise search for companies falls within the "less than 1% for everybody else" category I mentioned in regards to search. The revenue of Elastic Search is less than 0.5% of Google's revenue.