Hi,
Everytime I run Claude 3.7 with ontology edits it fails. Seems to be a bug. I think the model usually works with string outputs.
Best,
Jack
Hi Jack, thanks for flagging!
I’ve found and fixed the issue, and pushed the fix onto your stack. Can you please try again?
The problem was that Anthropic Claude models don’t allow multiple SYSTEM role messages in their chat API, whereas ~all other model providers do. We had a workaround being applied to a list of Claude models, but didn’t have the new Claude 3.7 on it yet. To fix, instead of just adding 3.7 to the list and having the same trouble when Claude $next comes out, I’ve replaced the model list with a model family list, so everything from Anthropic Claude should have this role munging fix going forward.
This is one of (many) model-specific quirks we need to deal with to make all the models in the model selector work seamlessly, and sometimes something slips through like happened here.
Thanks for helping make the product better!