Michał Moskal
Michał Moskal
I noticed you guys forked a bunch of controller code from AICI for your constraints. I think you might be interested in https://github.com/microsoft/llguidance - it implements a more general constraint...
[Docs in this repo](https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/prompt_format.md#model-response-format-5) say that when using custom tool calling with JSON one should use `Environment: ipython` and expect `{ "type": "function", "name": ...` [Docs on the website](https://www.llama.com/docs/model-cards-and-prompt-formats/llama3_1/#json-based-tool-calling) say...
This is a little experiment, to possibly allow using structured outputs with TypeChat. Note that you would want to only use this as a grammar for constraining the output, not...
With the following schema, xgrammar will accept `{"a": "wrong"}` even though it shouldn't ("a" is supposed to be an integer) ```json { "type": "object", "properties": { "a": {"type": "integer"} },...
Right now, we only return the error message. We should return the full response, and the error, so the user sees what was generated before we've seen the error. I...