Skip to content

grammars: cache decoded token codepoints for faster sampling #6811

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 19 commits into from
Closed
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
31 changes: 28 additions & 3 deletions llama.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -2368,6 +2368,11 @@ struct llama_context {

// control vectors
struct llama_control_vector cvec;

// caching token pieces & their decoded codepoints.
std::mutex token_cache_mutex;
std::vector<std::pair<std::vector<uint32_t>, llama_partial_utf8>>
token_codepoints_without_partial_utf8_prefix;
};

static size_t llama_get_device_count(const llama_model & model) {
Expand Down Expand Up @@ -14496,9 +14501,24 @@ void llama_sample_grammar(struct llama_context * ctx, llama_token_data_array * c
}
}

std::vector<std::pair<std::vector<uint32_t>, llama_partial_utf8>> candidates_decoded;
candidates_decoded.reserve(candidates->size);
{
// cache tokens & their decoded codepoints (for common case where there's no partial utf8 prefix bytes) for grammar-constrained sampling.
std::unique_lock<std::mutex> lock(ctx->token_cache_mutex);
if (ctx->token_codepoints_without_partial_utf8_prefix.empty()) {
auto n_vocab = llama_n_vocab(llama_get_model(ctx));
ctx->token_codepoints_without_partial_utf8_prefix.resize(n_vocab);
for (llama_token id = 0; id < n_vocab; ++id) {
const std::string & piece = ctx->model.vocab.cache_token_to_piece.at(id);
ctx->token_codepoints_without_partial_utf8_prefix[id] = decode_utf8(piece, {0, 0});
}
}
}

// Store decoded codepoints when they are not cached (happens when there's a partial utf8 string prefix).
std::vector<std::pair<std::vector<uint32_t>, llama_partial_utf8>> candidates_decoded;
if (grammar->partial_utf8.n_remain > 0) {
candidates_decoded.reserve(candidates->size);
}
std::vector<llama_grammar_candidate> candidates_grammar;
candidates_grammar.reserve(candidates->size);

Expand All @@ -14512,6 +14532,9 @@ void llama_sample_grammar(struct llama_context * ctx, llama_token_data_array * c
}
} else if (piece.empty() || piece[0] == 0) {
candidates->data[i].logit = -INFINITY;
} else if (grammar->partial_utf8.n_remain == 0){
const auto & decoded = ctx->token_codepoints_without_partial_utf8_prefix.at(id);
candidates_grammar.push_back({ i, decoded.first.data(), decoded.second });
} else {
candidates_decoded.push_back(decode_utf8(piece, grammar->partial_utf8));
candidates_grammar.push_back({ i, candidates_decoded.back().first.data(), candidates_decoded.back().second });
Expand Down Expand Up @@ -14707,7 +14730,9 @@ void llama_grammar_accept_token(struct llama_context * ctx, struct llama_grammar
const std::string & piece = ctx->model.vocab.cache_token_to_piece.at(token);

// Note terminating 0 in decoded string
const auto decoded = decode_utf8(piece, grammar->partial_utf8);
const auto decoded = grammar->partial_utf8.n_remain == 0
? ctx->token_codepoints_without_partial_utf8_prefix[token]
: decode_utf8(piece, grammar->partial_utf8);
const auto & code_points = decoded.first;
std::vector<std::vector<const llama_grammar_element *>> tmp_new_stacks;
for (auto it = code_points.begin(), end = code_points.end() - 1; it != end; ++it) {
Expand Down
Loading