TIS OpenAI frontend, make trust_remote_code configurable

[chorus-over-flanger]

Is your feature request related to a problem? Please describe.

For OpenAI frontend of TIS currently it's not directly possible to load and serve models from huggingface-hub without transformers builtin model architectures, like Bespoke-MiniCheck-7B . Happening because In tokenizer.py trust_remote_code defaults to False which is a reasonable safety measure, but not configurable on per-model basis

Describe the solution you'd like

Suggesting to update the codebase in order make trust_remote_code configurable on per-model basis

Describe alternatives you've considered

Patched tokenizer.py by setting trust_remote_code default to True 🥲

Additional context

No additional context provided