Looks like there will be a premium non-open source model for mistral, hopefully it’s the 180B version and the lower param models will remain open source.
This article mentions “Mistral-7B-V01” and “Mistral-7B-Instruct-V01” under “New Models in Models as a Service (MaaS)”.
Being Premium doesn’t mean it won’t be open source, it may just mean that it’s a branded, supported, and/or packaged version.
And Meta LLAMA will be offered as a service.
That makes total sense. I know we all love open source but honestly, is there any open source business model for LLMs? OSS companies offer commercial services but does this make sense for models? I doubt it. I guess Meta is making a single penny with Llama.
Os models with an ecosystem and support could be a reasonable business. Barrier to entry is expensive though alas.
Or time gated costs. Sub for the most up to date model so we can keep investing in rnd and then release the older ones fully later.
They have a premium 70b model that they have shown investors and enterprise customers.
They will likely opensource everything except the strongest model they have similar to what phind does.
I just hope we can still access the 70b model through the azure api.
They did already show a 70B model at some event few days ago https://techcrunch.com/2023/11/09/theres-something-going-on-with-ai-startups-in-france/ so maybe that’s the premium one??
Ooh, open or not, I’m seriously excited about this. Given the performance of Mistral-7B, a Mistral-180B model made available over an API of some sort would have a serious chance of dethroning GPT-4.
Sad.