4 ms·
For some more context see my submission from a few days ago - https://news.ycombinator.com/item?id=39175611 https://news.ycombinator.com/item?id=39175611, altho
by Tiberium 3y ago
For some more context see my submission from a few days ago - https://news.ycombinator.com/item?id=39175611 https://news.ycombinator.com/item?id=39175611, although it's admittedly not mistral-medium, but llama2 trained on the same dataset (since the outputs do match quite often with the API)
- whimsicalism 3y agohow do we know it’s not mistral medium
- wut42 3y agoIt has been confirmed by the Mistral CEO: > To quickly start working with a few selected customers, we retrained this model from Llama 2 the minute we got access to our entire cluster — the pretraining finished on the day of Mistral 7B release.
- whimsicalism 3y agoThat does not confirm that it is not Mistral Medium. It is written to give that impression, but if it weren’t Mistral Medium, I would expect them to explicitly say so.
- wut42 3y agoI think it does, as Mistral usually do their own models, and also, they couldn't fully commercialise Mistral Medium if it was LLama 2 based.
- whimsicalism 3y agoGood point. Likely Llama trained on similar corpora to what Mistral medium uses
- _boffin_ 3y agoWhat do you mean? Didn’t llama2’s tos say something about if you had something like 600m users at that current moment in time, you can’t use it? If you don’t, you’re good to go.