CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-03 09:22:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

parallel : fix crash when -n -1

Browse Source
This commit is contained in:
Georgi Gerganov
2023-09-28 15:48:38 +03:00
parent 25856900db
commit 4ad0676927
1 changed files with 3 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
examples/parallel/parallel.cpp
View File

@@ -114,7 +114,7 @@ int main(int argc, char ** argv) {
for (size_t i = 0; i < clients.size(); ++i) {
auto & client = clients[i];
client.id = i;
client.tokens_prev.resize(params.n_predict);
client.tokens_prev.resize(std::max(256, params.n_predict));
std::fill(client.tokens_prev.begin(), client.tokens_prev.end(), 0);
}
@@ -321,7 +321,8 @@ int main(int argc, char ** argv) {
// client.id, client.seq_id, id, client.n_decoded, client.i_batch, token_str.c_str());
if (client.n_decoded > 2 &&
(id == llama_token_eos(ctx) || client.n_decoded + client.n_prompt >= params.n_predict ||
(id == llama_token_eos(ctx) ||
(params.n_predict > 0 && client.n_decoded + client.n_prompt >= params.n_predict) ||
client.response.find("User:") != std::string::npos ||
client.response.find('\n') != std::string::npos)) {
// basic reverse prompt