diff --git a/src/llama-grammar.cpp b/src/llama-grammar.cpp index 75d5d750c3..7682e2a404 100644 --- a/src/llama-grammar.cpp +++ b/src/llama-grammar.cpp @@ -1343,6 +1343,7 @@ void llama_grammar_accept_impl(struct llama_grammar & grammar, llama_token token auto constrained_str = grammar.trigger_buffer.substr(start); grammar.trigger_buffer.clear(); + llama_grammar_accept_token(grammar, -1, constrained_str); grammar.trigger_buffer_positions.clear(); LLAMA_LOG_DEBUG("Grammar triggered on regex: '%s'\n", constrained_str.c_str()); return;