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

server : fix slot selection by lru, migrate lcs to size_t

Browse source
This commit is contained in:
sasha0552 2024-11-01 17:52:07 +00:00 • committed by GitHub
parent ba6f62eb79
commit aa4277cf97
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 8 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

1
examples/server/server.cpp
View file

@ -247,6 +247,7 @@ struct server_slot {
if (is_processing()) {
SLT_INF(*this, "stop processing: n_past = %d, truncated = %d\n", n_past, truncated);
t_last_used = ggml_time_us();
t_token_generation = (ggml_time_us() - t_start_generation) / 1e3;
state = SLOT_STATE_IDLE;
callback_on_release(id);

14
examples/server/utils.hpp
View file

@ -453,20 +453,20 @@ static size_t longest_common_subsequence(const llama_tokens & a, const llama_tok
}
// get the lengths of the input sequences
int a_len = a.size();
int b_len = b.size();
size_t a_len = a.size();
size_t b_len = b.size();
// initialize the maximum length of the longest common subsequence (LCS)
int max_length = 0;
size_t max_length = 0;
// use two rows instead of a 2D matrix to optimize space
std::vector<int> prev_row(b_len + 1, 0);
std::vector<int> curr_row(b_len + 1, 0);
std::vector<size_t> prev_row(b_len + 1, 0);
std::vector<size_t> curr_row(b_len + 1, 0);
// iterate through the elements of a
for (int i = 1; i <= a_len; i++) {
for (size_t i = 1; i <= a_len; i++) {
// iterate through the elements of b
for (int j = 1; j <= b_len; j++) {
for (size_t j = 1; j <= b_len; j++) {
// if elements at the current positions match
if (a[i - 1] == b[j - 1]) {
// if it's the first element of either sequences, set LCS length to 1