As you might’ve noticed, Llama-3 405 is out, and so is Qwen-Long when running on our Helmholtz-WestAI hardware. This is because the JURECA system is scheduled for maintenance tomorrow, and it’s not accepting long time jobs now. Sorry for the inconvenience. There are good news, though: there are two models right at the top of the list on the web ui: Phi and Qwen-Long, just running on the newer hardware. Phi is, at least for today, the default one on the list, so most people using the web ui will be using it, and will notice that this thing is FAST. I put it as the default to stress test the infrastructure. - Pros: We are testing the production instance of Blablador The Next Generation (R) - Cons: - We are testing, things come and go (the testing on the test instance is done - we’re going live, baby) - API does not show all the new models yet - Random glitches The infra is not complete yet, but soon one will be able to use it on the api, and use all the resources newer models allow: function calling, image support, etc. Hope you enjoy it! Let’s bark! Alex