diff --git a/main.cpp b/main.cpp index ee0952f74..c96f9edc9 100644 --- a/main.cpp +++ b/main.cpp @@ -85,7 +85,7 @@ struct llama_model { // load the model's weights from a file bool llama_model_load(const std::string & fname, llama_model & model, gpt_vocab & vocab, int n_ctx) { - printf("%s: loading model from '%s' - please wait ...\n", __func__, fname.c_str()); + fprintf(stderr, "%s: loading model from '%s' - please wait ...\n", __func__, fname.c_str()); std::vector f_buf(1024*1024); @@ -127,16 +127,16 @@ bool llama_model_load(const std::string & fname, llama_model & model, gpt_vocab n_ff = ((2*(4*hparams.n_embd)/3 + hparams.n_mult - 1)/hparams.n_mult)*hparams.n_mult; n_parts = LLAMA_N_PARTS.at(hparams.n_embd); - printf("%s: n_vocab = %d\n", __func__, hparams.n_vocab); - printf("%s: n_ctx = %d\n", __func__, hparams.n_ctx); - printf("%s: n_embd = %d\n", __func__, hparams.n_embd); - printf("%s: n_mult = %d\n", __func__, hparams.n_mult); - printf("%s: n_head = %d\n", __func__, hparams.n_head); - printf("%s: n_layer = %d\n", __func__, hparams.n_layer); - printf("%s: n_rot = %d\n", __func__, hparams.n_rot); - printf("%s: f16 = %d\n", __func__, hparams.f16); - printf("%s: n_ff = %d\n", __func__, n_ff); - printf("%s: n_parts = %d\n", __func__, n_parts); + fprintf(stderr, "%s: n_vocab = %d\n", __func__, hparams.n_vocab); + fprintf(stderr, "%s: n_ctx = %d\n", __func__, hparams.n_ctx); + fprintf(stderr, "%s: n_embd = %d\n", __func__, hparams.n_embd); + fprintf(stderr, "%s: n_mult = %d\n", __func__, hparams.n_mult); + fprintf(stderr, "%s: n_head = %d\n", __func__, hparams.n_head); + fprintf(stderr, "%s: n_layer = %d\n", __func__, hparams.n_layer); + fprintf(stderr, "%s: n_rot = %d\n", __func__, hparams.n_rot); + fprintf(stderr, "%s: f16 = %d\n", __func__, hparams.f16); + fprintf(stderr, "%s: n_ff = %d\n", __func__, n_ff); + fprintf(stderr, "%s: n_parts = %d\n", __func__, n_parts); } // load vocab @@ -161,7 +161,7 @@ bool llama_model_load(const std::string & fname, llama_model & model, gpt_vocab vocab.id_to_token[i] = word; //if (i < 30000) { - // printf("%s: vocab[%d] = '%s'\n", __func__, i, word.c_str()); + // fprintf(stderr, "%s: vocab[%d] = '%s'\n", __func__, i, word.c_str()); //} } } @@ -220,7 +220,7 @@ bool llama_model_load(const std::string & fname, llama_model & model, gpt_vocab ctx_size += (5 + 10*n_layer)*256; // object overhead - printf("%s: ggml ctx size = %6.2f MB\n", __func__, ctx_size/(1024.0*1024.0)); + fprintf(stderr, "%s: ggml ctx size = %6.2f MB\n", __func__, ctx_size/(1024.0*1024.0)); } // create the ggml context @@ -307,7 +307,7 @@ bool llama_model_load(const std::string & fname, llama_model & model, gpt_vocab const size_t memory_size = ggml_nbytes(model.memory_k) + ggml_nbytes(model.memory_v); - printf("%s: memory_size = %8.2f MB, n_mem = %d\n", __func__, memory_size/1024.0/1024.0, n_mem); + fprintf(stderr, "%s: memory_size = %8.2f MB, n_mem = %d\n", __func__, memory_size/1024.0/1024.0, n_mem); } const size_t file_offset = fin.tellg(); @@ -325,7 +325,7 @@ bool llama_model_load(const std::string & fname, llama_model & model, gpt_vocab fname_part += "." + std::to_string(i); } - printf("%s: loading model part %d/%d from '%s'\n", __func__, i+1, n_parts, fname_part.c_str()); + fprintf(stderr, "%s: loading model part %d/%d from '%s'\n", __func__, i+1, n_parts, fname_part.c_str()); fin = std::ifstream(fname_part, std::ios::binary); fin.rdbuf()->pubsetbuf(f_buf.data(), f_buf.size()); @@ -336,7 +336,7 @@ bool llama_model_load(const std::string & fname, llama_model & model, gpt_vocab int n_tensors = 0; size_t total_size = 0; - printf("%s: ", __func__); + fprintf(stderr, "%s: ", __func__); while (true) { int32_t n_dims; @@ -436,7 +436,7 @@ bool llama_model_load(const std::string & fname, llama_model & model, gpt_vocab if (0) { static const char * ftype_str[] = { "f32", "f16", "q4_0", "q4_1", }; - printf("%24s - [%5d, %5d], type = %6s, split = %d\n", name.data(), ne[0], ne[1], ftype_str[ftype], split_type); + fprintf(stderr, "%24s - [%5d, %5d], type = %6s, split = %d\n", name.data(), ne[0], ne[1], ftype_str[ftype], split_type); } size_t bpe = 0; @@ -499,16 +499,16 @@ bool llama_model_load(const std::string & fname, llama_model & model, gpt_vocab total_size += ggml_nbytes(tensor)/n_parts; } - //printf("%42s - [%5d, %5d], type = %6s, %6.2f MB\n", name.data(), ne[0], ne[1], ftype == 0 ? "float" : "f16", ggml_nbytes(tensor)/1024.0/1024.0); + //fprintf(stderr, "%42s - [%5d, %5d], type = %6s, %6.2f MB\n", name.data(), ne[0], ne[1], ftype == 0 ? "float" : "f16", ggml_nbytes(tensor)/1024.0/1024.0); if (++n_tensors % 8 == 0) { - printf("."); - fflush(stdout); + fprintf(stderr, "."); + fflush(stderr); } } - printf(" done\n"); + fprintf(stderr, " done\n"); - printf("%s: model size = %8.2f MB / num tensors = %d\n", __func__, total_size/1024.0/1024.0, n_tensors); + fprintf(stderr, "%s: model size = %8.2f MB / num tensors = %d\n", __func__, total_size/1024.0/1024.0, n_tensors); } fin.close(); @@ -552,7 +552,7 @@ bool llama_eval( if (mem_per_token > 0 && mem_per_token*N > buf_size) { const size_t buf_size_new = 1.1*(mem_per_token*N); // add 10% to account for ggml object overhead - //printf("\n%s: reallocating buffer from %zu to %zu bytes\n", __func__, buf_size, buf_size_new); + //fprintf(stderr, "\n%s: reallocating buffer from %zu to %zu bytes\n", __func__, buf_size, buf_size_new); // reallocate buf_size = buf_size_new; @@ -744,7 +744,7 @@ bool llama_eval( if (mem_per_token == 0) { mem_per_token = ggml_used_mem(ctx0)/N; } - //printf("used_mem = %zu\n", ggml_used_mem(ctx0)); + //fprintf(stderr, "used_mem = %zu\n", ggml_used_mem(ctx0)); ggml_free(ctx0); @@ -780,7 +780,7 @@ int main(int argc, char ** argv) { params.seed = time(NULL); } - printf("%s: seed = %d\n", __func__, params.seed); + fprintf(stderr, "%s: seed = %d\n", __func__, params.seed); std::mt19937 rng(params.seed); if (params.prompt.empty()) { @@ -822,13 +822,13 @@ int main(int argc, char ** argv) { // tokenize the reverse prompt std::vector antiprompt_inp = ::llama_tokenize(vocab, params.antiprompt, false); - printf("\n"); - printf("%s: prompt: '%s'\n", __func__, params.prompt.c_str()); - printf("%s: number of tokens in prompt = %zu\n", __func__, embd_inp.size()); + 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++) { - printf("%6d -> '%s'\n", embd_inp[i], vocab.id_to_token.at(embd_inp[i]).c_str()); + fprintf(stderr, "%6d -> '%s'\n", embd_inp[i], vocab.id_to_token.at(embd_inp[i]).c_str()); } - printf("\n"); + fprintf(stderr, "\n"); if (params.interactive) { #if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__)) struct sigaction sigint_action; @@ -838,19 +838,19 @@ int main(int argc, char ** argv) { sigaction(SIGINT, &sigint_action, NULL); #endif - printf("%s: interactive mode on.\n", __func__); + fprintf(stderr, "%s: interactive mode on.\n", __func__); if(antiprompt_inp.size()) { - printf("%s: reverse prompt: '%s'\n", __func__, params.antiprompt.c_str()); - printf("%s: number of tokens in reverse prompt = %zu\n", __func__, antiprompt_inp.size()); + fprintf(stderr, "%s: reverse prompt: '%s'\n", __func__, params.antiprompt.c_str()); + fprintf(stderr, "%s: number of tokens in reverse prompt = %zu\n", __func__, antiprompt_inp.size()); for (int i = 0; i < (int) antiprompt_inp.size(); i++) { - printf("%6d -> '%s'\n", antiprompt_inp[i], vocab.id_to_token.at(antiprompt_inp[i]).c_str()); + fprintf(stderr, "%6d -> '%s'\n", antiprompt_inp[i], vocab.id_to_token.at(antiprompt_inp[i]).c_str()); } - printf("\n"); + fprintf(stderr, "\n"); } } - printf("sampling parameters: temp = %f, top_k = %d, top_p = %f, repeat_last_n = %i, repeat_penalty = %f\n", params.temp, params.top_k, params.top_p, params.repeat_last_n, params.repeat_penalty); - printf("\n\n"); + fprintf(stderr, "sampling parameters: temp = %f, top_k = %d, top_p = %f, repeat_last_n = %i, repeat_penalty = %f\n", params.temp, params.top_k, params.top_p, params.repeat_last_n, params.repeat_penalty); + fprintf(stderr, "\n\n"); std::vector embd; @@ -864,7 +864,7 @@ int main(int argc, char ** argv) { if (params.interactive) { - printf("== Running in interactive mode. ==\n" + fprintf(stderr, "== Running in interactive mode. ==\n" #if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__)) " - Press Ctrl+C to interject at any time.\n" #endif @@ -892,7 +892,7 @@ int main(int argc, char ** argv) { const int64_t t_start_us = ggml_time_us(); if (!llama_eval(model, params.n_threads, n_past, embd, logits, mem_per_token)) { - printf("Failed to predict\n"); + fprintf(stderr, "Failed to predict\n"); return 1; } @@ -1005,7 +1005,7 @@ int main(int argc, char ** argv) { // end of text token if (embd.back() == 2) { - printf(" [end of text]\n"); + fprintf(stderr, " [end of text]\n"); break; } } @@ -1015,12 +1015,12 @@ int main(int argc, char ** argv) { { const int64_t t_main_end_us = ggml_time_us(); - printf("\n\n"); - printf("%s: mem per token = %8zu bytes\n", __func__, mem_per_token); - printf("%s: load time = %8.2f ms\n", __func__, t_load_us/1000.0f); - printf("%s: sample time = %8.2f ms\n", __func__, t_sample_us/1000.0f); - printf("%s: predict time = %8.2f ms / %.2f ms per token\n", __func__, t_predict_us/1000.0f, t_predict_us/1000.0f/n_past); - printf("%s: total time = %8.2f ms\n", __func__, (t_main_end_us - t_main_start_us)/1000.0f); + fprintf(stderr, "\n\n"); + fprintf(stderr, "%s: mem per token = %8zu bytes\n", __func__, mem_per_token); + fprintf(stderr, "%s: load time = %8.2f ms\n", __func__, t_load_us/1000.0f); + fprintf(stderr, "%s: sample time = %8.2f ms\n", __func__, t_sample_us/1000.0f); + fprintf(stderr, "%s: predict time = %8.2f ms / %.2f ms per token\n", __func__, t_predict_us/1000.0f, t_predict_us/1000.0f/n_past); + fprintf(stderr, "%s: total time = %8.2f ms\n", __func__, (t_main_end_us - t_main_start_us)/1000.0f); } ggml_free(model.ctx);