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

server : normalize embeddings (#5956)

Browse source 
* output normalize embedding in '/v1/embeddings'

* common : reuse llama_embd_normalize

* common : better normalize impl

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
SeungWon Jeong 2024-03-09 21:27:58 +09:00 • committed by GitHub
parent 2c4f566c88
commit fb215c3832
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 30 additions and 14 deletions
Download patch file Download diff file Expand all files Collapse all files

8
examples/server/server.cpp
View file

@ -1327,6 +1327,8 @@ struct server_context {
const int n_embd = llama_n_embd(model);
std::vector<float> embd_res(n_embd, 0.0f);
for (int i = 0; i < batch.n_tokens; ++i) {
if (!batch.logits[i] || batch.seq_id[i][0] != slot.id + 1) {
continue;
@ -1350,8 +1352,10 @@ struct server_context {
continue;
}
llama_embd_normalize(embd, embd_res.data(), n_embd);
res.data = json {
{"embedding", std::vector<float>(embd, embd + n_embd)},
{"embedding", embd_res},
};
}
@ -3354,6 +3358,8 @@ int main(int argc, char ** argv) {
// get the result
server_task_result result = ctx_server.queue_results.recv(id_task);
ctx_server.queue_results.remove_waiting_task_id(id_task);
// append to the responses
responses.push_back(result.data);
}