grammars: early exit when no next_candidates to reject
grammars: cache decoded tokens
grammars: faster llama_grammar_copy
grammars: fix bad merge
grammars: keep llama_grammar_copy non-quadratic optim for later
grammars: move token caches to llama_context
grammars: cache codepoints in llama_new_context_with_model
grammar: nit (layout)
grammars: nits (revert const grammar sig, fix comment)
Update llama.cpp
Co-authored-by: Clint Herron <hanclinto@gmail.com>
grammars: mutex-guarded lazy caching of token pieces in llama_sample_grammar
grammars: remove early exit --> https://github.com/ggerganov/llama.cpp/pull/7370
