Update AI models

[PierreMesure]

Here are some updates regarding the AI model list:

• New Claude 3.5 Sonnet (October 2024)
• New Claude 3.5 Haiku
• New Qwen 2.5 14B (although I'm not sure you're offering it on your infrastructure? If not, you should, it's miles better than Qwen 1.5)
• multilingual-e5-large-instruct is almost 5% better than the non-instruct version on the Scandinavian Embedding Benchmark

I didn't remove any of the old models but Claude 3.5 Sonnet (June 2024) should probably be deprecated as Anthropic might stop supporting it.

[Jontpan]

I think we should point the Claude models to the latest snapshot (-latest).

I know that they discourage this (to ensure consistent behaviour) but I think that by and large, users will not notice a difference. This would save us some maintenance headaches, and would also mirror how we have set it up with the OpenAI models.

I am ready to be challenged on this however, and would be great to get the public sector perspective!

What do you think @PierreMesure ?

[PierreMesure]

That sounds good to me, I have changed the Claude models to -latest when it was available.

I didn't want to change the existing list of models because I don't know if that file could affect existing assistants for some of your customers. In the long run, maybe you could fetch model lists directly from the providers? Librechat does it if fetch:true is set at provider level. So you would need to change your structure a little bit.