| 8 | #endif |
| 9 | |
| 10 | int main(int argc, char ** argv) { |
| 11 | gpt_params params; |
| 12 | |
| 13 | if (!gpt_params_parse(argc, argv, params)) { |
| 14 | return 1; |
| 15 | } |
| 16 | |
| 17 | params.embedding = true; |
| 18 | |
| 19 | print_build_info(); |
| 20 | |
| 21 | if (params.seed == LLAMA_DEFAULT_SEED) { |
| 22 | params.seed = time(NULL); |
| 23 | } |
| 24 | |
| 25 | fprintf(stderr, "%s: seed = %u\n", __func__, params.seed); |
| 26 | |
| 27 | std::mt19937 rng(params.seed); |
| 28 | if (params.random_prompt) { |
| 29 | params.prompt = gpt_random_prompt(rng); |
| 30 | } |
| 31 | |
| 32 | llama_backend_init(params.numa); |
| 33 | |
| 34 | llama_model * model; |
| 35 | llama_context * ctx; |
| 36 | |
| 37 | // load the model |
| 38 | std::tie(model, ctx) = llama_init_from_gpt_params(params); |
| 39 | if (model == NULL) { |
| 40 | fprintf(stderr, "%s: error: unable to load model\n", __func__); |
| 41 | return 1; |
| 42 | } |
| 43 | |
| 44 | const int n_ctx_train = llama_n_ctx_train(model); |
| 45 | const int n_ctx = llama_n_ctx(ctx); |
| 46 | |
| 47 | if (n_ctx > n_ctx_train) { |
| 48 | fprintf(stderr, "%s: warning: model was trained on only %d context tokens (%d specified)\n", |
| 49 | __func__, n_ctx_train, n_ctx); |
| 50 | } |
| 51 | |
| 52 | // print system information |
| 53 | { |
| 54 | fprintf(stderr, "\n"); |
| 55 | fprintf(stderr, "%s\n", get_system_info(params).c_str()); |
| 56 | } |
| 57 | |
| 58 | int n_past = 0; |
| 59 | |
| 60 | // tokenize the prompt |
| 61 | auto embd_inp = ::llama_tokenize(ctx, params.prompt, true); |
| 62 | |
| 63 | if (params.verbose_prompt) { |
| 64 | fprintf(stderr, "\n"); |
| 65 | fprintf(stderr, "%s: prompt: '%s'\n", __func__, params.prompt.c_str()); |
| 66 | fprintf(stderr, "%s: number of tokens in prompt = %zu\n", __func__, embd_inp.size()); |
| 67 | for (int i = 0; i < (int) embd_inp.size(); i++) { |
nothing calls this directly
no test coverage detected