diff --git a/src/main.cpp b/src/main.cpp index c6ef84c..0b6c8e8 100644 --- a/src/main.cpp +++ b/src/main.cpp @@ -60,6 +60,11 @@ int main(int argc, char **argv) .default_value(0.1f) .scan<'g', float>(); + program.add_argument("-b", "--batch-size") + .help("set batch size for model completion") + .default_value(512) + .scan<'i',int>(); + program.add_argument("prompt").remaining(); @@ -96,6 +101,7 @@ int main(int argc, char **argv) config.n_threads = program.get("--threads"); config.temp = program.get("--temperature"); config.top_p = program.get("--top-p"); + config.n_batch = program.get("--batch-size"); if(model_type.compare("codegen") == 0) { spdlog::info("Initializing GPT-J type model for '{}' model", model_type); @@ -131,6 +137,7 @@ int main(int argc, char **argv) return "Hello world"; }); + CROW_ROUTE(app, "/copilot_internal/v2/token")([](){ //return "Hello world";