Bogdan Buduroiu

`is_valid_input` allows inputs that don't match function signature type

1

As seen in [this skipped pytest unit test](https://github.com/aurelio-labs/semantic-router/pull/96/files#diff-c443490a143d8b29a394eba3b857e9e432529b0eab4fbb3eb4139a748188a9dbR29-R37), `is_valid_input` will consider `{"timezone": None}` a valid input for function signature `(timezone: str) -> str` Origin of bug: https://github.com/aurelio-labs/semantic-router/blob/666b361e9d3bc24eae831060e5de114d073d39b1/semantic_router/utils/function_call.py#L104-L108

Re-raise LLM exceptions instead of raising new exception

4

In Python, re-raising exceptions (`raise` instead of `raise Exception(e)` preserves the original stack trace, helping with debugging. Change the LLM `__call__` exception handling to re-raise errors happening during generation https://github.com/aurelio-labs/semantic-router/blob/f8fcd43a7018d6216668250ff99d9acbe7f16789/semantic_router/llms/openai.py#L53

Apple has released [MLX](https://github.com/ml-explore/mlx) as an Apple silicon optimised Array framework. Users have observed up to 2x faster inference times for examples problems such as MNIST ([Twitter](https://twitter.com/mrdbourke/status/1732925761299136517))

Hi, I'm running Chat UI container inside of Kubernetes, with mongo. The entire svelte build can't complete and my pods get `OOMKilled` as Chat UI seems to almost reach 4GB...