vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

grammars: reserve rejects & next candidates

Browse source
This commit is contained in:
Olivier Chafik 2024-04-10 23:05:14 +01:00
parent f4183afe6a
commit 3732ad9c22
1 changed files with 3 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

3
llama.cpp
View file

@ -11951,6 +11951,7 @@ static std::vector<llama_grammar_candidate> llama_grammar_reject_candidates_for_
const std::vector<llama_grammar_candidate> & candidates) { const std::vector<llama_grammar_candidate> & candidates) {
std::vector<llama_grammar_candidate> rejects; std::vector<llama_grammar_candidate> rejects;
rejects.reserve(candidates.size());
if (stack.empty()) { if (stack.empty()) {
for (const auto & tok : candidates) { for (const auto & tok : candidates) {
@ -11964,6 +11965,8 @@ static std::vector<llama_grammar_candidate> llama_grammar_reject_candidates_for_
const llama_grammar_element * stack_pos = stack.back(); const llama_grammar_element * stack_pos = stack.back();
std::vector<llama_grammar_candidate> next_candidates; std::vector<llama_grammar_candidate> next_candidates;
next_candidates.reserve(candidates.size());
for (const auto & tok : candidates) { for (const auto & tok : candidates) {
if (*tok.code_points == 0) { if (*tok.code_points == 0) {
// reached end of full codepoints in token, reject iff it ended in a partial sequence // reached end of full codepoints in token, reject iff it ended in a partial sequence