From 8fe813130a4bc7dc9447ebc1f7393c18bc53e84c Mon Sep 17 00:00:00 2001 From: Branden Butler Date: Mon, 25 Sep 2023 17:42:41 -0500 Subject: [PATCH] Update MPI example to follow main changes --- examples/mpi/mpi.cpp | 416 ++++++++++++++++++++----------------------- ggml-mpi.c | 2 - 2 files changed, 195 insertions(+), 223 deletions(-) diff --git a/examples/mpi/mpi.cpp b/examples/mpi/mpi.cpp index 393ef1b2a..84f15a82d 100644 --- a/examples/mpi/mpi.cpp +++ b/examples/mpi/mpi.cpp @@ -1,9 +1,5 @@ -// Defines sigaction on msys: -#ifndef _GNU_SOURCE -#define _GNU_SOURCE -#endif - #include "common.h" + #include "console.h" #include "llama.h" #include "build-info.h" @@ -40,7 +36,6 @@ #pragma warning(disable: 4244 4267) // possible loss of data #endif - static llama_context ** g_ctx; static llama_model ** g_model; static gpt_params * g_params; @@ -49,10 +44,12 @@ static std::ostringstream * g_output_ss; static std::vector * g_output_tokens; static bool is_interacting = false; -void write_logfile( - const llama_context * ctx, const gpt_params & params, const llama_model * model, - const std::vector input_tokens, const std::string output, const std::vector output_tokens) { +static void write_logfile( + const llama_context * ctx, const gpt_params & params, const llama_model * model, + const std::vector & input_tokens, const std::string & output, + const std::vector & output_tokens +) { if (params.logdir.empty()) { return; } @@ -93,7 +90,7 @@ void write_logfile( } #if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__)) || defined (_WIN32) -void sigint_handler(int signo) { +static void sigint_handler(int signo) { if (signo == SIGINT) { if (!is_interacting) { is_interacting = true; @@ -109,7 +106,6 @@ void sigint_handler(int signo) { #endif int main(int argc, char ** argv) { - gpt_params params; g_params = ¶ms; @@ -156,6 +152,15 @@ int main(int argc, char ** argv) { } wordfree(&splitOptions); +#ifndef LOG_DISABLE_LOGS + log_set_target(log_filename_generator("main", "log")); + LOG_TEE("Log start\n"); + log_dump_cmdline(argc, argv); +#endif // LOG_DISABLE_LOGS + + // TODO: Dump params ? + //LOG("Params perplexity: %s\n", LOG_TOSTR(params.perplexity)); + // save choice to use color for later // (note for later: this is a slightly awkward choice) console::init(params.simple_io, params.use_color); @@ -178,34 +183,28 @@ int main(int argc, char ** argv) { } if (params.rope_freq_base != 10000.0) { - fprintf(stderr, "%s: warning: changing RoPE frequency base to %g (default 10000.0)\n", __func__, params.rope_freq_base); + LOG_TEE("%s: warning: changing RoPE frequency base to %g (default 10000.0)\n", __func__, params.rope_freq_base); } if (params.rope_freq_scale != 1.0) { - fprintf(stderr, "%s: warning: scaling RoPE frequency by %g (default 1.0)\n", __func__, params.rope_freq_scale); + LOG_TEE("%s: warning: scaling RoPE frequency by %g (default 1.0)\n", __func__, params.rope_freq_scale); } - if (params.n_ctx > 2048) { - // TODO: determine the actual max context of the model (e.g. 4096 for LLaMA v2) and use that instead of 2048 - fprintf(stderr, "%s: warning: base model only supports context sizes no greater than 2048 tokens (%d specified)\n", __func__, params.n_ctx); - } else if (params.n_ctx < 8) { - fprintf(stderr, "%s: warning: minimum context size is 8, using minimum size.\n", __func__); - params.n_ctx = 8; - } - - fprintf(stderr, "%s: build = %d (%s)\n", __func__, BUILD_NUMBER, BUILD_COMMIT); + LOG_TEE("%s: build = %d (%s)\n", __func__, BUILD_NUMBER, BUILD_COMMIT); + LOG_TEE("%s: built with %s for %s\n", __func__, BUILD_COMPILER, BUILD_TARGET); if (params.seed == LLAMA_DEFAULT_SEED) { params.seed = time(NULL); } - fprintf(stderr, "%s: seed = %u\n", __func__, params.seed); + LOG_TEE("%s: seed = %u\n", __func__, params.seed); std::mt19937 rng(params.seed); if (params.random_prompt) { params.prompt = gpt_random_prompt(rng); } + LOG("%s: llama backend init\n", __func__); llama_backend_init(params.numa); llama_model * model; @@ -215,6 +214,7 @@ int main(int argc, char ** argv) { g_ctx = &ctx; // load the model and apply lora adapter, if any + LOG("%s: load the model and apply lora adapter, if any\n", __func__); std::tie(model, ctx) = llama_init_from_gpt_params(params); if (params.cfg_scale > 1.f) { struct llama_context_params lparams = llama_context_params_from_gpt_params(params); @@ -222,14 +222,23 @@ int main(int argc, char ** argv) { } if (model == NULL) { - fprintf(stderr, "%s: error: unable to load model\n", __func__); + LOG_TEE("%s: error: unable to load model\n", __func__); return 1; } + const int n_ctx_train = llama_n_ctx_train(ctx); + if (params.n_ctx > n_ctx_train) { + LOG_TEE("%s: warning: model was trained on only %d context tokens (%d specified)\n", + __func__, n_ctx_train, params.n_ctx); + } else if (params.n_ctx < 8) { + LOG_TEE("%s: warning: minimum context size is 8, using minimum size.\n", __func__); + params.n_ctx = 8; + } + // print system information { - fprintf(stderr, "\n"); - fprintf(stderr, "system_info: n_threads = %d / %d | %s\n", + LOG_TEE("\n"); + LOG_TEE("system_info: n_threads = %d / %d | %s\n", params.n_threads, std::thread::hardware_concurrency(), llama_print_system_info()); } @@ -241,13 +250,14 @@ int main(int argc, char ** argv) { return 0; } + llama_split_layers_weighted(ctx, params.mpi_layer_split); std::string path_session = params.path_prompt_cache; std::vector session_tokens; if (!path_session.empty()) { - fprintf(stderr, "%s: attempting to load saved session from '%s'\n", __func__, path_session.c_str()); + LOG_TEE("%s: attempting to load saved session from '%s'\n", __func__, path_session.c_str()); // fopen to check for existing session FILE * fp = std::fopen(path_session.c_str(), "rb"); @@ -257,33 +267,38 @@ int main(int argc, char ** argv) { session_tokens.resize(params.n_ctx); size_t n_token_count_out = 0; if (!llama_load_session_file(ctx, path_session.c_str(), session_tokens.data(), session_tokens.capacity(), &n_token_count_out)) { - fprintf(stderr, "%s: error: failed to load session file '%s'\n", __func__, path_session.c_str()); + LOG_TEE("%s: error: failed to load session file '%s'\n", __func__, path_session.c_str()); return 1; } session_tokens.resize(n_token_count_out); llama_set_rng_seed(ctx, params.seed); - fprintf(stderr, "%s: loaded a session with prompt size of %d tokens\n", __func__, (int) session_tokens.size()); + LOG_TEE("%s: loaded a session with prompt size of %d tokens\n", __func__, (int) session_tokens.size()); } else { - fprintf(stderr, "%s: session file does not exist, will create\n", __func__); + LOG_TEE("%s: session file does not exist, will create\n", __func__); } } - // Add BOS if SPM tokenizer const bool add_bos = llama_vocab_type(ctx) == LLAMA_VOCAB_TYPE_SPM; + LOG("add_bos: %d\n", add_bos); - // tokenize the prompt std::vector embd_inp; if (params.interactive_first || params.instruct || !params.prompt.empty() || session_tokens.empty()) { + LOG("tokenize the prompt\n"); embd_inp = ::llama_tokenize(ctx, params.prompt, add_bos); } else { + LOG("use session tokens\n"); embd_inp = session_tokens; } + LOG("prompt: \"%s\"\n", log_tostr(params.prompt)); + LOG("tokens: %s\n", LOG_TOKENS_TOSTR_PRETTY(ctx, embd_inp)); + // Should not run without any tokens if (embd_inp.empty()) { embd_inp.push_back(llama_token_bos(ctx)); + LOG("embd_inp was considered empty and bos was added: %s\n", LOG_TOKENS_TOSTR_PRETTY(ctx, embd_inp)); } // Tokenize negative prompt @@ -291,23 +306,31 @@ int main(int argc, char ** argv) { int guidance_offset = 0; int original_prompt_len = 0; if (ctx_guidance) { + LOG("cfg_negative_prompt: \"%s\"\n", log_tostr(params.cfg_negative_prompt)); + guidance_inp = ::llama_tokenize(ctx_guidance, params.cfg_negative_prompt, add_bos); + LOG("guidance_inp tokenized: %s\n", LOG_TOKENS_TOSTR_PRETTY(ctx_guidance, guidance_inp)); std::vector original_inp = ::llama_tokenize(ctx, params.prompt, add_bos); + LOG("original_inp tokenized: %s\n", LOG_TOKENS_TOSTR_PRETTY(ctx, original_inp)); + original_prompt_len = original_inp.size(); guidance_offset = (int)guidance_inp.size() - original_prompt_len; + LOG("original_prompt_len: %s", log_tostr(original_prompt_len)); + LOG("guidance_offset: %s", log_tostr(guidance_offset)); } const int n_ctx = llama_n_ctx(ctx); + LOG("n_ctx: %d\n", n_ctx); if ((int) embd_inp.size() > n_ctx - 4) { - fprintf(stderr, "%s: error: prompt is too long (%d tokens, max %d)\n", __func__, (int) embd_inp.size(), n_ctx - 4); + LOG_TEE("%s: error: prompt is too long (%d tokens, max %d)\n", __func__, (int) embd_inp.size(), n_ctx - 4); return 1; } // debug message about similarity of saved session, if applicable size_t n_matching_session_tokens = 0; - if (session_tokens.size()) { + if (!session_tokens.empty()) { for (llama_token id : session_tokens) { if (n_matching_session_tokens >= embd_inp.size() || id != embd_inp[n_matching_session_tokens]) { break; @@ -315,22 +338,27 @@ int main(int argc, char ** argv) { n_matching_session_tokens++; } if (params.prompt.empty() && n_matching_session_tokens == embd_inp.size()) { - fprintf(stderr, "%s: using full prompt from session file\n", __func__); + LOG_TEE("%s: using full prompt from session file\n", __func__); } else if (n_matching_session_tokens >= embd_inp.size()) { - fprintf(stderr, "%s: session file has exact match for prompt!\n", __func__); + LOG_TEE("%s: session file has exact match for prompt!\n", __func__); } else if (n_matching_session_tokens < (embd_inp.size() / 2)) { - fprintf(stderr, "%s: warning: session file has low similarity to prompt (%zu / %zu tokens); will mostly be reevaluated\n", - __func__, n_matching_session_tokens, embd_inp.size()); + LOG_TEE("%s: warning: session file has low similarity to prompt (%zu / %zu tokens); will mostly be reevaluated\n", + __func__, n_matching_session_tokens, embd_inp.size()); } else { - fprintf(stderr, "%s: session file matches %zu / %zu tokens of prompt\n", - __func__, n_matching_session_tokens, embd_inp.size()); + LOG_TEE("%s: session file matches %zu / %zu tokens of prompt\n", + __func__, n_matching_session_tokens, embd_inp.size()); } } + LOGLN( + "recalculate the cached logits (check): embd_inp.empty() %s, n_matching_session_tokens %zu, embd_inp.size() %zu, session_tokens.size() %zu, embd_inp.size() %zu", + log_tostr(embd_inp.empty()), n_matching_session_tokens, embd_inp.size(), session_tokens.size(), embd_inp.size()); + // if we will use the cache for the full prompt without reaching the end of the cache, force // reevaluation of the last token token to recalculate the cached logits - if (!embd_inp.empty() && n_matching_session_tokens == embd_inp.size() && - session_tokens.size() > embd_inp.size()) { + if (!embd_inp.empty() && n_matching_session_tokens == embd_inp.size() && session_tokens.size() > embd_inp.size()) { + LOGLN("recalculate the cached logits (do): session_tokens.resize( %zu )", embd_inp.size() - 1); + session_tokens.resize(embd_inp.size() - 1); } @@ -343,6 +371,9 @@ int main(int argc, char ** argv) { const auto inp_pfx = ::llama_tokenize(ctx, "\n\n### Instruction:\n\n", add_bos); const auto inp_sfx = ::llama_tokenize(ctx, "\n\n### Response:\n\n", false); + LOG("inp_pfx: %s\n", LOG_TOKENS_TOSTR_PRETTY(ctx, inp_pfx)); + LOG("inp_sfx: %s\n", LOG_TOKENS_TOSTR_PRETTY(ctx, inp_sfx)); + // in instruct mode, we inject a prefix and a suffix to each input by the user if (params.instruct) { params.interactive_first = true; @@ -355,30 +386,30 @@ int main(int argc, char ** argv) { } 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()); + LOG_TEE("\n"); + LOG_TEE("%s: prompt: '%s'\n", __func__, params.prompt.c_str()); + LOG_TEE("%s: number of tokens in prompt = %zu\n", __func__, embd_inp.size()); for (int i = 0; i < (int) embd_inp.size(); i++) { - fprintf(stderr, "%6d -> '%s'\n", embd_inp[i], llama_token_to_piece(ctx, embd_inp[i]).c_str()); + LOG_TEE("%6d -> '%s'\n", embd_inp[i], llama_token_to_piece(ctx, embd_inp[i]).c_str()); } if (ctx_guidance) { - fprintf(stderr, "\n"); - fprintf(stderr, "%s: negative prompt: '%s'\n", __func__, params.cfg_negative_prompt.c_str()); - fprintf(stderr, "%s: number of tokens in negative prompt = %zu\n", __func__, guidance_inp.size()); + LOG_TEE("\n"); + LOG_TEE("%s: negative prompt: '%s'\n", __func__, params.cfg_negative_prompt.c_str()); + LOG_TEE("%s: number of tokens in negative prompt = %zu\n", __func__, guidance_inp.size()); for (int i = 0; i < (int) guidance_inp.size(); i++) { - fprintf(stderr, "%6d -> '%s'\n", guidance_inp[i], llama_token_to_piece(ctx, guidance_inp[i]).c_str()); + LOG_TEE("%6d -> '%s'\n", guidance_inp[i], llama_token_to_piece(ctx, guidance_inp[i]).c_str()); } } if (params.n_keep > 0) { - fprintf(stderr, "%s: static prompt based on n_keep: '", __func__); + LOG_TEE("%s: static prompt based on n_keep: '", __func__); for (int i = 0; i < params.n_keep; i++) { - fprintf(stderr, "%s", llama_token_to_piece(ctx, embd_inp[i]).c_str()); + LOG_TEE("%s", llama_token_to_piece(ctx, embd_inp[i]).c_str()); } - fprintf(stderr, "'\n"); + LOG_TEE("'\n"); } - fprintf(stderr, "\n"); + LOG_TEE("\n"); } if (params.interactive) { @@ -395,58 +426,59 @@ int main(int argc, char ** argv) { SetConsoleCtrlHandler(reinterpret_cast(console_ctrl_handler), true); #endif - fprintf(stderr, "%s: interactive mode on.\n", __func__); + LOG_TEE("%s: interactive mode on.\n", __func__); - if (params.antiprompt.size()) { - for (auto antiprompt : params.antiprompt) { - fprintf(stderr, "Reverse prompt: '%s'\n", antiprompt.c_str()); + if (!params.antiprompt.empty()) { + for (const auto & antiprompt : params.antiprompt) { + LOG_TEE("Reverse prompt: '%s'\n", antiprompt.c_str()); } } if (params.input_prefix_bos) { - fprintf(stderr, "Input prefix with BOS\n"); + LOG_TEE("Input prefix with BOS\n"); } if (!params.input_prefix.empty()) { - fprintf(stderr, "Input prefix: '%s'\n", params.input_prefix.c_str()); + LOG_TEE("Input prefix: '%s'\n", params.input_prefix.c_str()); } if (!params.input_suffix.empty()) { - fprintf(stderr, "Input suffix: '%s'\n", params.input_suffix.c_str()); + LOG_TEE("Input suffix: '%s'\n", params.input_suffix.c_str()); } } - fprintf(stderr, "sampling: repeat_last_n = %d, repeat_penalty = %f, presence_penalty = %f, frequency_penalty = %f, top_k = %d, tfs_z = %f, top_p = %f, typical_p = %f, temp = %f, mirostat = %d, mirostat_lr = %f, mirostat_ent = %f\n", + LOG_TEE("sampling: repeat_last_n = %d, repeat_penalty = %f, presence_penalty = %f, frequency_penalty = %f, top_k = %d, tfs_z = %f, top_p = %f, typical_p = %f, temp = %f, mirostat = %d, mirostat_lr = %f, mirostat_ent = %f\n", params.repeat_last_n, params.repeat_penalty, params.presence_penalty, params.frequency_penalty, params.top_k, params.tfs_z, params.top_p, params.typical_p, params.temp, params.mirostat, params.mirostat_eta, params.mirostat_tau); - fprintf(stderr, "generate: n_ctx = %d, n_batch = %d, n_predict = %d, n_keep = %d\n", n_ctx, params.n_batch, params.n_predict, params.n_keep); - fprintf(stderr, "\n\n"); + LOG_TEE("generate: n_ctx = %d, n_batch = %d, n_predict = %d, n_keep = %d\n", n_ctx, params.n_batch, params.n_predict, params.n_keep); + LOG_TEE("\n\n"); + struct llama_grammar * grammar = NULL; grammar_parser::parse_state parsed_grammar; - llama_grammar * grammar = NULL; + if (!params.grammar.empty()) { parsed_grammar = grammar_parser::parse(params.grammar.c_str()); // will be empty (default) if there are parse errors if (parsed_grammar.rules.empty()) { return 1; } - fprintf(stderr, "%s: grammar:\n", __func__); + LOG_TEE("%s: grammar:\n", __func__); grammar_parser::print_grammar(stderr, parsed_grammar); - fprintf(stderr, "\n"); + LOG_TEE("\n"); { auto it = params.logit_bias.find(llama_token_eos(ctx)); if (it != params.logit_bias.end() && it->second == -INFINITY) { - fprintf(stderr, "%s: warning: EOS token is disabled, which will cause most grammars to fail\n", __func__); + LOG_TEE("%s: warning: EOS token is disabled, which will cause most grammars to fail\n", __func__); } } std::vector grammar_rules(parsed_grammar.c_rules()); grammar = llama_grammar_init( - grammar_rules.data(), grammar_rules.size(), parsed_grammar.symbol_ids.at("root")); + grammar_rules.data(), grammar_rules.size(), parsed_grammar.symbol_ids.at("root")); } // TODO: replace with ring-buffer - std::vector last_n_tokens(n_ctx); - std::fill(last_n_tokens.begin(), last_n_tokens.end(), 0); + std::vector last_tokens(n_ctx); + std::fill(last_tokens.begin(), last_tokens.end(), 0); if (params.interactive) { const char *control_message; @@ -458,11 +490,11 @@ int main(int argc, char ** argv) { " - To return control without starting a new line, end your input with '/'.\n" " - If you want to submit another line, end your input with '\\'.\n"; } - fprintf(stderr, "== Running in interactive mode. ==\n" + LOG_TEE("== Running in interactive mode. ==\n"); #if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__)) || defined (_WIN32) - " - Press Ctrl+C to interject at any time.\n" + LOG_TEE( " - Press Ctrl+C to interject at any time.\n"); #endif - "%s\n", control_message); + LOG_TEE( "%s\n", control_message); is_interacting = params.interactive_first; } @@ -487,27 +519,27 @@ int main(int argc, char ** argv) { std::vector embd; std::vector embd_guidance; - // do one empty run to warm up the model - { - const std::vector tmp = { llama_token_bos(ctx), }; - llama_eval(ctx, tmp.data(), tmp.size(), 0, params.n_threads); - llama_reset_timings(ctx); - } + const int n_vocab = llama_n_vocab(ctx); + + std::vector candidates; + candidates.reserve(n_vocab); while ((n_remain != 0 && !is_antiprompt) || params.interactive) { // predict - if (embd.size() > 0) { + if (!embd.empty()) { // Note: n_ctx - 4 here is to match the logic for commandline prompt handling via // --prompt or --file which uses the same value. - auto max_embd_size = n_ctx - 4; + int max_embd_size = n_ctx - 4; + // Ensure the input doesn't exceed the context size by truncating embd if necessary. - if ((int)embd.size() > max_embd_size) { - auto skipped_tokens = embd.size() - max_embd_size; + if ((int) embd.size() > max_embd_size) { + const int skipped_tokens = (int) embd.size() - max_embd_size; + embd.resize(max_embd_size); + console::set_display(console::error); - printf("<>", skipped_tokens, skipped_tokens != 1 ? "s" : ""); + printf("<>", skipped_tokens, skipped_tokens != 1 ? "s" : ""); console::set_display(console::reset); fflush(stdout); - embd.resize(max_embd_size); } // infinite text generation via context swapping @@ -516,28 +548,26 @@ int main(int argc, char ** argv) { // - take half of the last (n_ctx - n_keep) tokens and recompute the logits in batches if (n_past + (int) embd.size() + std::max(0, guidance_offset) > n_ctx) { if (params.n_predict == -2) { - fprintf(stderr, "\n\n%s: context full, stopping generation\n", __func__); + LOG_TEE("\n\n%s: context full and n_predict == -%d => stopping\n", __func__, params.n_predict); break; } const int n_left = n_past - params.n_keep; + LOG("context full, swapping: n_past = %d, n_left = %d, n_ctx = %d, n_keep = %d\n", n_past, n_left, n_ctx, params.n_keep); + // always keep the first token - BOS - n_past = std::max(1, params.n_keep); + n_past = std::max(1, params.n_keep); n_past_guidance = std::max(1, params.n_keep + guidance_offset); - // insert n_left/2 tokens at the start of embd from last_n_tokens - embd.insert(embd.begin(), last_n_tokens.begin() + n_ctx - n_left/2 - embd.size(), last_n_tokens.end() - embd.size()); + LOG("after swap: n_past = %d, n_past_guidance = %d\n", n_past, n_past_guidance); - // stop saving session if we run out of context + // insert n_left/2 tokens at the start of embd from last_tokens + embd.insert(embd.begin(), last_tokens.begin() + n_ctx - n_left/2 - embd.size(), last_tokens.end() - embd.size()); + + LOG("embd: %s\n", LOG_TOKENS_TOSTR_PRETTY(ctx, embd)); + + LOG("clear session path\n"); path_session.clear(); - - //printf("\n---\n"); - //printf("resetting: '"); - //for (int i = 0; i < (int) embd.size(); i++) { - // printf("%s", llama_token_to_piece(ctx, embd[i])); - //} - //printf("'\n"); - //printf("\n---\n"); } // try to reuse a matching prefix from the loaded session instead of re-eval (via n_past) @@ -567,7 +597,7 @@ int main(int argc, char ** argv) { if (ctx_guidance) { int input_size = 0; - llama_token* input_buf = NULL; + llama_token * input_buf = NULL; if (n_past_guidance < (int) guidance_inp.size()) { // Guidance context should have the same data with these modifications: @@ -577,28 +607,25 @@ int main(int argc, char ** argv) { embd_guidance = guidance_inp; if (embd.begin() + original_prompt_len < embd.end()) { embd_guidance.insert( - embd_guidance.end(), - embd.begin() + original_prompt_len, - embd.end() + embd_guidance.end(), + embd.begin() + original_prompt_len, + embd.end() ); } - input_buf = embd_guidance.data(); + input_buf = embd_guidance.data(); input_size = embd_guidance.size(); - //fprintf(stderr, "\n---------------------\n"); - //for (int i = 0; i < (int) embd_guidance.size(); i++) { - //fprintf(stderr, "%s", llama_token_to_piece(ctx, embd_guidance[i])); - //} - //fprintf(stderr, "\n---------------------\n"); + + LOG("guidance context: %s\n", LOG_TOKENS_TOSTR_PRETTY(ctx, embd_guidance)); } else { - input_buf = embd.data(); + input_buf = embd.data(); input_size = embd.size(); } for (int i = 0; i < input_size; i += params.n_batch) { int n_eval = std::min(input_size - i, params.n_batch); if (llama_eval(ctx_guidance, input_buf + i, n_eval, n_past_guidance, params.n_threads)) { - fprintf(stderr, "%s : failed to eval\n", __func__); + LOG_TEE("%s : failed to eval\n", __func__); return 1; } @@ -611,14 +638,20 @@ int main(int argc, char ** argv) { if (n_eval > params.n_batch) { n_eval = params.n_batch; } + + LOG("eval: %s\n", LOG_TOKENS_TOSTR_PRETTY(ctx, embd)); + if (llama_eval(ctx, &embd[i], n_eval, n_past, params.n_threads)) { - fprintf(stderr, "%s : failed to eval\n", __func__); + LOG_TEE("%s : failed to eval\n", __func__); return 1; } + n_past += n_eval; + + LOG("n_past = %d\n", n_past); } - if (embd.size() > 0 && !path_session.empty()) { + if (!embd.empty() && !path_session.empty()) { session_tokens.insert(session_tokens.end(), embd.begin(), embd.end()); n_session_consumed = session_tokens.size(); } @@ -628,106 +661,21 @@ int main(int argc, char ** argv) { embd_guidance.clear(); if ((int) embd_inp.size() <= n_consumed && !is_interacting) { - // out of user input, sample next token - const float temp = params.temp; - const int32_t top_k = params.top_k <= 0 ? llama_n_vocab(ctx) : params.top_k; - const float top_p = params.top_p; - const float tfs_z = params.tfs_z; - const float typical_p = params.typical_p; - const int32_t repeat_last_n = params.repeat_last_n < 0 ? n_ctx : params.repeat_last_n; - const float repeat_penalty = params.repeat_penalty; - const float alpha_presence = params.presence_penalty; - const float alpha_frequency = params.frequency_penalty; - const int mirostat = params.mirostat; - const float mirostat_tau = params.mirostat_tau; - const float mirostat_eta = params.mirostat_eta; - const bool penalize_nl = params.penalize_nl; - // optionally save the session on first sample (for faster prompt loading next time) if (!path_session.empty() && need_to_save_session && !params.prompt_cache_ro) { need_to_save_session = false; llama_save_session_file(ctx, path_session.c_str(), session_tokens.data(), session_tokens.size()); + + LOG("saved session to %s\n", path_session.c_str()); } - llama_token id = 0; + const llama_token id = llama_sample_token(ctx, ctx_guidance, grammar, params, last_tokens, candidates); - { - auto logits = llama_get_logits(ctx); - auto n_vocab = llama_n_vocab(ctx); + last_tokens.erase(last_tokens.begin()); + last_tokens.push_back(id); - // Apply params.logit_bias map - for (auto it = params.logit_bias.begin(); it != params.logit_bias.end(); it++) { - logits[it->first] += it->second; - } + LOG("last: %s\n", LOG_TOKENS_TOSTR_PRETTY(ctx, last_tokens)); - std::vector candidates; - candidates.reserve(n_vocab); - for (llama_token token_id = 0; token_id < n_vocab; token_id++) { - candidates.emplace_back(llama_token_data{token_id, logits[token_id], 0.0f}); - } - - llama_token_data_array candidates_p = { candidates.data(), candidates.size(), false }; - - if (ctx_guidance) { - llama_sample_classifier_free_guidance(ctx, &candidates_p, ctx_guidance, params.cfg_scale); - } - - // Apply penalties - float nl_logit = logits[llama_token_nl(ctx)]; - auto last_n_repeat = std::min(std::min((int)last_n_tokens.size(), repeat_last_n), n_ctx); - llama_sample_repetition_penalty(ctx, &candidates_p, - last_n_tokens.data() + last_n_tokens.size() - last_n_repeat, - last_n_repeat, repeat_penalty); - llama_sample_frequency_and_presence_penalties(ctx, &candidates_p, - last_n_tokens.data() + last_n_tokens.size() - last_n_repeat, - last_n_repeat, alpha_frequency, alpha_presence); - if (!penalize_nl) { - for (size_t idx = 0; idx < candidates_p.size; idx++) { - if (candidates_p.data[idx].id == llama_token_nl(ctx)) { - candidates_p.data[idx].logit = nl_logit; - break; - } - } - } - - if (grammar != NULL) { - llama_sample_grammar(ctx, &candidates_p, grammar); - } - - if (temp <= 0) { - // Greedy sampling - id = llama_sample_token_greedy(ctx, &candidates_p); - } else { - if (mirostat == 1) { - static float mirostat_mu = 2.0f * mirostat_tau; - const int mirostat_m = 100; - llama_sample_temperature(ctx, &candidates_p, temp); - id = llama_sample_token_mirostat(ctx, &candidates_p, mirostat_tau, mirostat_eta, mirostat_m, &mirostat_mu); - } else if (mirostat == 2) { - static float mirostat_mu = 2.0f * mirostat_tau; - llama_sample_temperature(ctx, &candidates_p, temp); - id = llama_sample_token_mirostat_v2(ctx, &candidates_p, mirostat_tau, mirostat_eta, &mirostat_mu); - } else { - // Temperature sampling - llama_sample_top_k(ctx, &candidates_p, top_k, 1); - llama_sample_tail_free(ctx, &candidates_p, tfs_z, 1); - llama_sample_typical(ctx, &candidates_p, typical_p, 1); - llama_sample_top_p(ctx, &candidates_p, top_p, 1); - llama_sample_temperature(ctx, &candidates_p, temp); - id = llama_sample_token(ctx, &candidates_p); - } - } - // printf("`%d`", candidates_p.size); - - if (grammar != NULL) { - llama_grammar_accept_token(ctx, grammar, id); - } - - last_n_tokens.erase(last_n_tokens.begin()); - last_n_tokens.push_back(id); - } - - // add it to the context embd.push_back(id); // echo this to console @@ -735,12 +683,15 @@ int main(int argc, char ** argv) { // decrement remaining sampling budget --n_remain; + + LOG("n_remain: %d\n", n_remain); } else { // some user input remains from prompt or interaction, forward it to processing + LOG("embd_inp.size(): %d, n_consumed: %d\n", (int) embd_inp.size(), n_consumed); while ((int) embd_inp.size() > n_consumed) { embd.push_back(embd_inp[n_consumed]); - last_n_tokens.erase(last_n_tokens.begin()); - last_n_tokens.push_back(embd_inp[n_consumed]); + last_tokens.erase(last_tokens.begin()); + last_tokens.push_back(embd_inp[n_consumed]); ++n_consumed; if ((int) embd.size() >= params.n_batch) { break; @@ -764,17 +715,16 @@ int main(int argc, char ** argv) { fflush(stdout); } // reset color to default if we there is no pending user input - if (input_echo && (int)embd_inp.size() == n_consumed) { + if (input_echo && (int) embd_inp.size() == n_consumed) { console::set_display(console::reset); } // if not currently processing queued inputs; if ((int) embd_inp.size() <= n_consumed) { - // check for reverse prompt - if (params.antiprompt.size()) { + if (!params.antiprompt.empty()) { std::string last_output; - for (auto id : last_n_tokens) { + for (auto id : last_tokens) { last_output += llama_token_to_piece(ctx, id); } @@ -785,10 +735,10 @@ int main(int argc, char ** argv) { for (std::string & antiprompt : params.antiprompt) { size_t extra_padding = params.interactive ? 0 : 2; size_t search_start_pos = last_output.length() > static_cast(antiprompt.length() + extra_padding) - ? last_output.length() - static_cast(antiprompt.length() + extra_padding) - : 0; + ? last_output.length() - static_cast(antiprompt.length() + extra_padding) + : 0; - if (last_output.find(antiprompt.c_str(), search_start_pos) != std::string::npos) { + if (last_output.find(antiprompt, search_start_pos) != std::string::npos) { if (params.interactive) { is_interacting = true; console::set_display(console::user_input); @@ -798,12 +748,18 @@ int main(int argc, char ** argv) { break; } } + + if (is_antiprompt) { + LOG("found antiprompt: %s\n", last_output.c_str()); + } } // deal with end of text token in interactive mode - if (last_n_tokens.back() == llama_token_eos(ctx)) { + if (last_tokens.back() == llama_token_eos(ctx)) { + LOG("found EOS token\n"); + if (params.interactive) { - if (params.antiprompt.size() != 0) { + if (!params.antiprompt.empty()) { // tokenize and inject first reverse prompt const auto first_antiprompt = ::llama_tokenize(ctx, params.antiprompt.front(), false); embd_inp.insert(embd_inp.end(), first_antiprompt.begin(), first_antiprompt.end()); @@ -820,16 +776,20 @@ int main(int argc, char ** argv) { } if (n_past > 0 && is_interacting) { + LOG("waiting for user input\n"); + if (params.instruct) { printf("\n> "); } if (params.input_prefix_bos) { + LOG("adding input prefix BOS token\n"); embd_inp.push_back(llama_token_bos(ctx)); } std::string buffer; if (!params.input_prefix.empty()) { + LOG("appending input prefix: '%s'\n", params.input_prefix.c_str()); buffer += params.input_prefix; printf("%s", buffer.c_str()); } @@ -849,23 +809,30 @@ int main(int argc, char ** argv) { if (buffer.length() > 1) { // append input suffix if any if (!params.input_suffix.empty()) { + LOG("appending input suffix: '%s'\n", params.input_suffix.c_str()); buffer += params.input_suffix; printf("%s", params.input_suffix.c_str()); } + LOG("buffer: '%s'\n", buffer.c_str()); + const size_t original_size = embd_inp.size(); // instruct mode: insert instruction prefix if (params.instruct && !is_antiprompt) { + LOG("inserting instruction prefix\n"); n_consumed = embd_inp.size(); embd_inp.insert(embd_inp.end(), inp_pfx.begin(), inp_pfx.end()); } - auto line_inp = ::llama_tokenize(ctx, buffer, false); + const auto line_inp = ::llama_tokenize(ctx, buffer, false); + LOG("input tokens: %s\n", LOG_TOKENS_TOSTR_PRETTY(ctx, line_inp)); + embd_inp.insert(embd_inp.end(), line_inp.begin(), line_inp.end()); // instruct mode: insert response suffix if (params.instruct) { + LOG("inserting instruction suffix\n"); embd_inp.insert(embd_inp.end(), inp_sfx.begin(), inp_sfx.end()); } @@ -876,6 +843,9 @@ int main(int argc, char ** argv) { } n_remain -= line_inp.size(); + LOG("n_remain: %d\n", n_remain); + } else { + LOG("empty line, passing control back\n"); } input_echo = false; // do not echo this again @@ -887,10 +857,10 @@ int main(int argc, char ** argv) { if (grammar != NULL) { llama_grammar_free(grammar); - std::vector grammar_rules( parsed_grammar.c_rules()); + std::vector grammar_rules(parsed_grammar.c_rules()); grammar = llama_grammar_init( - grammar_rules.data(), grammar_rules.size(), - parsed_grammar.symbol_ids.at("root")); + grammar_rules.data(), grammar_rules.size(), + parsed_grammar.symbol_ids.at("root")); } } is_interacting = false; @@ -899,7 +869,7 @@ int main(int argc, char ** argv) { // end of text token if (!embd.empty() && embd.back() == llama_token_eos(ctx) && !(params.instruct || params.interactive)) { - fprintf(stderr, " [end of text]\n"); + LOG_TEE(" [end of text]\n"); break; } @@ -912,7 +882,7 @@ int main(int argc, char ** argv) { } if (!path_session.empty() && params.prompt_cache_all && !params.prompt_cache_ro) { - fprintf(stderr, "\n%s: saving final output to session file '%s'\n", __func__, path_session.c_str()); + LOG_TEE("\n%s: saving final output to session file '%s'\n", __func__, path_session.c_str()); llama_save_session_file(ctx, path_session.c_str(), session_tokens.data(), session_tokens.size()); } @@ -928,5 +898,9 @@ int main(int argc, char ** argv) { } llama_backend_free(); +#ifndef LOG_DISABLE_LOGS + LOG_TEE("Log end\n") +#endif // LOG_DISABLE_LOGS + return 0; } diff --git a/ggml-mpi.c b/ggml-mpi.c index cef5ca6da..9217651d6 100644 --- a/ggml-mpi.c +++ b/ggml-mpi.c @@ -166,9 +166,7 @@ void ggml_mpi_scatter_layers( if (layer_ranges != NULL) { for (int i = 0; i < ctx_mpi->size * 2; i += 2) { - fprintf(stderr, "In iteration %d\n", i); flattened_ranges[i] = layer_ranges[i/2][0]; - fprintf(stderr, "Got first element\n"); flattened_ranges[i + 1] = layer_ranges[i/2][1]; } }