char-fork/llama.cpp

git clone https://git.t4t.associates/char-fork/llama.cpp

Georgi Gerganovtests : avoid building get-model.cpp many times (#26317)47f686f53

char/aliceai-t5-moe
brancheschar/aliceai-t5-moeall refs →
679 B22 linesraw
1// ref: https://github.com/ggml-org/llama.cpp/issues/4952#issuecomment-1892864763
2
3#include <thread>
4
5#include "llama.h"
6#include "common.h"
7
8// This creates a new context inside a pthread and then tries to exit cleanly.
9int main(int argc, char ** argv) {
10    auto * model_path = common_get_model_or_exit(argc, argv);
11
12    std::thread([&model_path]() {
13        llama_backend_init();
14        auto * model = llama_model_load_from_file(model_path, llama_model_default_params());
15        auto * ctx = llama_init_from_model(model, llama_context_default_params());
16        llama_free(ctx);
17        llama_model_free(model);
18        llama_backend_free();
19    }).join();
20
21    return 0;
22}