Module chat_template 

Render a model’s own chat template (the GGUF’s tokenizer.chat_template, Jinja) into a prompt. Pure, so it is tested without loading a model.

Using the model’s own template (instead of a generic one) is what makes the model answer in its trained format, and what lets callers pass template switches such as enable_thinking.

Structs

TemplateError

A template that failed to compile or render.

TemplateVars

Special tokens some templates reference ({{ bos_token }}).

Functions

merge_kwargs

Model defaults overlaid with the request’s own kwargs (request wins).

render_chat

Render messages with template, ending in the assistant turn. kwargs are extra template variables (e.g. enable_thinking).