mirror/LocalAI
1
0
Fork 0
mirror of https://github.com/mudler/LocalAI.git synced 2026-01-06 10:39:55 -06:00
Code Issues Packages Projects Releases Wiki Activity

feat(grpc): return consumed token count and update response accordingly (#2035)

Browse Source 
Fixes: #1920
This commit is contained in:
Ettore Di Giacinto
2024-04-15 19:47:11 +02:00
committed by GitHub
parent de3a1a0a8e
commit e843d7df0e
4 changed files with 20 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
backend/backend.proto
View File

@@ -114,6 +114,8 @@ message PredictOptions {
// The response message containing the result
message Reply {
bytes message = 1;
int32 tokens = 2;
int32 prompt_tokens = 3;
}
message ModelOptions {

8
backend/cpp/llama/grpc-server.cpp
View File

@@ -2332,6 +2332,10 @@ public:
std::string completion_text = result.result_json.value("content", "");
reply.set_message(completion_text);
int32_t tokens_predicted = result.result_json.value("tokens_predicted", 0);
reply.set_tokens(tokens_predicted);
int32_t tokens_evaluated = result.result_json.value("tokens_evaluated", 0);
reply.set_prompt_tokens(tokens_evaluated);
// Send the reply
writer->Write(reply);
@@ -2357,6 +2361,10 @@ public:
task_result result = llama.queue_results.recv(task_id);
if (!result.error && result.stop) {
completion_text = result.result_json.value("content", "");
int32_t tokens_predicted = result.result_json.value("tokens_predicted", 0);
int32_t tokens_evaluated = result.result_json.value("tokens_evaluated", 0);
reply->set_prompt_tokens(tokens_evaluated);
reply->set_tokens(tokens_predicted);
reply->set_message(completion_text);
}
else