2023-03-25 19:26:40 +01:00
|
|
|
#include "common.h"
|
|
|
|
#include "llama.h"
|
|
|
|
|
2023-04-16 12:13:00 +02:00
|
|
|
#include <ctime>
|
|
|
|
|
2023-06-16 20:23:53 +02:00
|
|
|
#if defined(_MSC_VER)
|
|
|
|
#pragma warning(disable: 4244 4267) // possible loss of data
|
|
|
|
#endif
|
|
|
|
|
2023-03-25 19:26:40 +01:00
|
|
|
int main(int argc, char ** argv) {
|
|
|
|
gpt_params params;
|
|
|
|
|
2023-09-07 19:22:29 +02:00
|
|
|
if (!gpt_params_parse(argc, argv, params)) {
|
2023-03-25 19:26:40 +01:00
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
params.embedding = true;
|
|
|
|
|
2023-09-15 22:59:49 +02:00
|
|
|
print_build_info();
|
2023-05-01 18:23:47 +02:00
|
|
|
|
2023-06-29 15:15:15 +02:00
|
|
|
if (params.seed == LLAMA_DEFAULT_SEED) {
|
2023-03-25 19:26:40 +01:00
|
|
|
params.seed = time(NULL);
|
|
|
|
}
|
|
|
|
|
2023-06-29 15:15:15 +02:00
|
|
|
fprintf(stderr, "%s: seed = %u\n", __func__, params.seed);
|
2023-03-25 19:26:40 +01:00
|
|
|
|
|
|
|
std::mt19937 rng(params.seed);
|
|
|
|
if (params.random_prompt) {
|
|
|
|
params.prompt = gpt_random_prompt(rng);
|
|
|
|
}
|
|
|
|
|
2023-07-10 17:49:56 +02:00
|
|
|
llama_backend_init(params.numa);
|
2023-05-20 10:06:11 +02:00
|
|
|
|
2023-06-24 10:47:58 +02:00
|
|
|
llama_model * model;
|
2023-03-25 19:26:40 +01:00
|
|
|
llama_context * ctx;
|
|
|
|
|
|
|
|
// load the model
|
2023-06-24 10:47:58 +02:00
|
|
|
std::tie(model, ctx) = llama_init_from_gpt_params(params);
|
|
|
|
if (model == NULL) {
|
2023-05-02 22:39:51 +02:00
|
|
|
fprintf(stderr, "%s: error: unable to load model\n", __func__);
|
|
|
|
return 1;
|
2023-03-25 19:26:40 +01:00
|
|
|
}
|
|
|
|
|
2023-09-28 21:42:38 +02:00
|
|
|
const int n_ctx_train = llama_n_ctx_train(model);
|
|
|
|
const int n_ctx = llama_n_ctx(ctx);
|
|
|
|
|
|
|
|
if (n_ctx > n_ctx_train) {
|
2023-09-08 17:43:35 +02:00
|
|
|
fprintf(stderr, "%s: warning: model was trained on only %d context tokens (%d specified)\n",
|
2023-09-28 21:42:38 +02:00
|
|
|
__func__, n_ctx_train, n_ctx);
|
2023-09-08 17:43:35 +02:00
|
|
|
}
|
|
|
|
|
2023-03-25 19:26:40 +01:00
|
|
|
// print system information
|
|
|
|
{
|
|
|
|
fprintf(stderr, "\n");
|
2023-09-28 21:42:38 +02:00
|
|
|
fprintf(stderr, "%s\n", get_system_info(params).c_str());
|
2023-03-25 19:26:40 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
int n_past = 0;
|
|
|
|
|
|
|
|
// tokenize the prompt
|
|
|
|
auto embd_inp = ::llama_tokenize(ctx, params.prompt, true);
|
|
|
|
|
|
|
|
if (params.verbose_prompt) {
|
|
|
|
fprintf(stderr, "\n");
|
|
|
|
fprintf(stderr, "%s: prompt: '%s'\n", __func__, params.prompt.c_str());
|
|
|
|
fprintf(stderr, "%s: number of tokens in prompt = %zu\n", __func__, embd_inp.size());
|
|
|
|
for (int i = 0; i < (int) embd_inp.size(); i++) {
|
2023-08-27 13:19:19 +02:00
|
|
|
fprintf(stderr, "%6d -> '%s'\n", embd_inp[i], llama_token_to_piece(ctx, embd_inp[i]).c_str());
|
2023-03-25 19:26:40 +01:00
|
|
|
}
|
|
|
|
fprintf(stderr, "\n");
|
|
|
|
}
|
|
|
|
|
2023-09-28 21:42:38 +02:00
|
|
|
if (embd_inp.size() > (size_t)n_ctx) {
|
2023-08-22 16:03:12 +02:00
|
|
|
fprintf(stderr, "%s: error: prompt is longer than the context window (%zu tokens, n_ctx = %d)\n",
|
2023-09-28 21:42:38 +02:00
|
|
|
__func__, embd_inp.size(), n_ctx);
|
2023-08-22 16:03:12 +02:00
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
while (!embd_inp.empty()) {
|
|
|
|
int n_tokens = std::min(params.n_batch, (int) embd_inp.size());
|
2023-09-28 21:42:38 +02:00
|
|
|
if (llama_decode(ctx, llama_batch_get_one(embd_inp.data(), n_tokens, n_past, 0))) {
|
2023-08-22 16:03:12 +02:00
|
|
|
fprintf(stderr, "%s : failed to eval\n", __func__);
|
|
|
|
return 1;
|
2023-03-25 19:26:40 +01:00
|
|
|
}
|
2023-08-22 16:03:12 +02:00
|
|
|
n_past += n_tokens;
|
|
|
|
embd_inp.erase(embd_inp.begin(), embd_inp.begin() + n_tokens);
|
|
|
|
}
|
2023-03-25 19:26:40 +01:00
|
|
|
|
2023-09-28 21:42:38 +02:00
|
|
|
const int n_embd = llama_n_embd(model);
|
|
|
|
const auto * embeddings = llama_get_embeddings(ctx);
|
2023-03-25 19:26:40 +01:00
|
|
|
|
2023-08-22 16:03:12 +02:00
|
|
|
for (int i = 0; i < n_embd; i++) {
|
|
|
|
printf("%f ", embeddings[i]);
|
2023-03-25 19:26:40 +01:00
|
|
|
}
|
2023-08-22 16:03:12 +02:00
|
|
|
printf("\n");
|
2023-03-25 19:26:40 +01:00
|
|
|
|
|
|
|
llama_print_timings(ctx);
|
|
|
|
llama_free(ctx);
|
2023-06-24 10:47:58 +02:00
|
|
|
llama_free_model(model);
|
2023-03-25 19:26:40 +01:00
|
|
|
|
2023-07-10 17:49:56 +02:00
|
|
|
llama_backend_free();
|
|
|
|
|
2023-03-25 19:26:40 +01:00
|
|
|
return 0;
|
|
|
|
}
|