Expand description
Local chat-template rendering.
The serving runtime’s own template is the preferred source (/apply-template
on llama.cpp), but runtimes such as vLLM expose no such endpoint. This module
renders the same Jinja template locally with minijinja.
llama.cpp renders with the C++ minijinja plus Python-style string methods,
which the Rust crate does not implement. Those methods are supplied through
Environment::set_unknown_method_callback so a template written for
transformers still renders here. python_string_methods mirrors CPython
semantics, including treating the argument of strip/lstrip/rstrip as a
set of characters rather than a prefix.
Structs§
- Local
Renderer - A chat template rendered in-process.