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

retrieval : fix memory leak in retrieval query handling (#8955)

Browse source 
* retrieval

* Reuse querybatch to reduce frequent memory allocation

* delete unused white space
This commit is contained in:
gtygo 2024-08-15 15:40:12 +08:00 • committed by GitHub
parent 37501d9c79
commit 4b9afbbe90
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 3 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

4
examples/retrieval/retrieval.cpp
View file

@ -253,6 +253,8 @@ int main(int argc, char ** argv) {
chunks[i].tokens.clear();
}
struct llama_batch query_batch = llama_batch_init(n_batch, 0, 1);
// start loop, receive query and return top k similar chunks based on cosine similarity
std::string query;
while (true) {
@ -260,7 +262,6 @@ int main(int argc, char ** argv) {
std::getline(std::cin, query);
std::vector<int32_t> query_tokens = llama_tokenize(ctx, query, true);
struct llama_batch query_batch = llama_batch_init(n_batch, 0, 1);
batch_add_seq(query_batch, query_tokens, 0);
std::vector<float> query_emb(n_embd, 0);
@ -293,6 +294,7 @@ int main(int argc, char ** argv) {
}
// clean up
llama_batch_free(query_batch);
llama_print_timings(ctx);
llama_free(ctx);
llama_free_model(model);