From 3a13d1e82924810bfc62026da54e0484ac4fa79e Mon Sep 17 00:00:00 2001 From: Henri Vasserman Date: Thu, 13 Jul 2023 20:24:31 +0300 Subject: [PATCH] Apply formatting. --- examples/server/server.cpp | 822 +++++++++++++++---------------------- 1 file changed, 327 insertions(+), 495 deletions(-) diff --git a/examples/server/server.cpp b/examples/server/server.cpp index 296c5d646..481804d54 100644 --- a/examples/server/server.cpp +++ b/examples/server/server.cpp @@ -22,8 +22,7 @@ using namespace httplib; using json = nlohmann::json; -struct server_params -{ +struct server_params { std::string hostname = "127.0.0.1"; std::string public_path = "examples/server/public"; int32_t port = 8080; @@ -32,10 +31,8 @@ struct server_params }; // completion token output with probabilities -struct completion_token_output -{ - struct token_prob - { +struct completion_token_output { + struct token_prob { llama_token tok; float prob; }; @@ -44,40 +41,30 @@ struct completion_token_output llama_token tok; }; -static size_t common_part(const std::vector &a, const std::vector &b) -{ +static size_t common_part(const std::vector & a, const std::vector & b) { size_t i; - for (i = 0; i < a.size() && i < b.size() && a[i] == b[i]; i++) - { - } + for (i = 0; i < a.size() && i < b.size() && a[i] == b[i]; i++) {} return i; } -enum stop_type -{ +enum stop_type { STOP_FULL, STOP_PARTIAL, }; -static bool ends_with(const std::string &str, const std::string &suffix) -{ +static bool ends_with(const std::string & str, const std::string & suffix) { return str.size() >= suffix.size() && - 0 == str.compare(str.size() - suffix.size(), suffix.size(), suffix); + 0 == str.compare(str.size() - suffix.size(), suffix.size(), suffix); } -static size_t find_partial_stop_string(const std::string &stop, - const std::string &text) -{ - if (!text.empty() && !stop.empty()) - { +static size_t find_partial_stop_string(const std::string & stop, + const std::string & text) { + if (!text.empty() && !stop.empty()) { const char text_last_char = text.back(); - for (int64_t char_index = stop.size() - 1; char_index >= 0; char_index--) - { - if (stop[char_index] == text_last_char) - { + for (int64_t char_index = stop.size() - 1; char_index >= 0; char_index--) { + if (stop[char_index] == text_last_char) { const std::string current_partial = stop.substr(0, char_index + 1); - if (ends_with(text, current_partial)) - { + if (ends_with(text, current_partial)) { return text.size() - char_index - 1; } } @@ -86,30 +73,26 @@ static size_t find_partial_stop_string(const std::string &stop, return std::string::npos; } -template -static std::string tokens_to_str(llama_context *ctx, Iter begin, Iter end) -{ +template +static std::string tokens_to_str(llama_context * ctx, Iter begin, Iter end) { std::string ret; - for (; begin != end; ++begin) - { + for (; begin != end; ++begin) { ret += llama_token_to_str(ctx, *begin); } return ret; } -static void server_log(const char *level, const char *function, int line, - const char *message, const nlohmann::ordered_json &extra) -{ - nlohmann::ordered_json log{ - {"timestamp", time(nullptr)}, - {"level", level}, - {"function", function}, - {"line", line}, - {"message", message}, +static void server_log(const char * level, const char * function, int line, + const char * message, const nlohmann::ordered_json & extra) { + nlohmann::ordered_json log { + { "timestamp", time(nullptr) }, + { "level", level }, + { "function", function }, + { "line", line }, + { "message", message }, }; - if (!extra.empty()) - { + if (!extra.empty()) { log.merge_patch(extra); } @@ -119,12 +102,10 @@ static void server_log(const char *level, const char *function, int line, } // format incomplete utf-8 multibyte character for output -static std::string tokens_to_output_formatted_string(const llama_context *ctx, const llama_token token) -{ +static std::string tokens_to_output_formatted_string(const llama_context * ctx, const llama_token token) { std::string out = token == -1 ? "" : llama_token_to_str(ctx, token); // if first bit is 1, meaning it's a partial character - if (out.size() > 0 && (out[0] & 0x80) == 0x80) - { + if (out.size() > 0 && (out[0] & 0x80) == 0x80) { std::stringstream ss; ss << std::hex << (out[0] & 0xff); std::string res(ss.str()); @@ -134,22 +115,19 @@ static std::string tokens_to_output_formatted_string(const llama_context *ctx, c } // convert a vector of completion_token_output to json -static json probs_vector_to_json(const llama_context *ctx, const std::vector probs) -{ +static json probs_vector_to_json(const llama_context * ctx, const std::vector probs) { json out = json::array(); - for (const auto &prob : probs) - { + for (const auto & prob : probs) { json probs_for_token = json::array(); - for (const auto &p : prob.probs) - { + for (const auto & p : prob.probs) { std::string tok_str = tokens_to_output_formatted_string(ctx, p.tok); - probs_for_token.push_back(json{ - {"tok_str", tok_str}, - {"prob", p.prob}, + probs_for_token.push_back(json { + { "tok_str", tok_str }, + { "prob", p.prob }, }); } std::string tok_str = tokens_to_output_formatted_string(ctx, prob.tok); - out.push_back(json{ + out.push_back(json { {"content", tok_str}, {"probs", probs_for_token}, }); @@ -160,13 +138,11 @@ static json probs_vector_to_json(const llama_context *ctx, const std::vector embd; std::vector last_n_tokens; - llama_model *model = nullptr; - llama_context *ctx = nullptr; + llama_model * model = nullptr; + llama_context * ctx = nullptr; gpt_params params; bool truncated = false; @@ -204,27 +179,22 @@ struct llama_server_context std::mutex mutex; - std::unique_lock lock() - { + std::unique_lock lock() { return std::unique_lock(mutex); } - ~llama_server_context() - { - if (ctx) - { + ~llama_server_context() { + if (ctx) { llama_free(ctx); ctx = nullptr; } - if (model) - { + if (model) { llama_free_model(model); model = nullptr; } } - void rewind() - { + void rewind() { params.antiprompt.clear(); num_prompt_tokens = 0; num_tokens_predicted = 0; @@ -242,13 +212,11 @@ struct llama_server_context n_past = 0; } - bool loadModel(const gpt_params ¶ms_) - { + bool loadModel(const gpt_params & params_) { params = params_; std::tie(model, ctx) = llama_init_from_gpt_params(params); - if (model == nullptr) - { - LOG_ERROR("unable to load model", {{"model", params_.model}}); + if (model == nullptr) { + LOG_ERROR("unable to load model", {{ "model", params_.model }}); return false; } @@ -257,21 +225,18 @@ struct llama_server_context return true; } - void loadPrompt() - { + void loadPrompt() { params.prompt.insert(0, 1, ' '); // always add a first space std::vector prompt_tokens = ::llama_tokenize(ctx, params.prompt, true); num_prompt_tokens = prompt_tokens.size(); - if (params.n_keep < 0) - { + if (params.n_keep < 0) { params.n_keep = (int)num_prompt_tokens; } params.n_keep = std::min(params.n_ctx - 4, params.n_keep); // if input prompt is too big, truncate like normal - if (num_prompt_tokens >= (size_t)params.n_ctx) - { + if (num_prompt_tokens>= (size_t)params.n_ctx) { const int n_left = (params.n_ctx - params.n_keep) / 2; std::vector new_tokens(prompt_tokens.begin(), prompt_tokens.begin() + params.n_keep); const int erased_blocks = (num_prompt_tokens - params.n_keep - n_left - 1) / n_left; @@ -279,17 +244,15 @@ struct llama_server_context std::copy(prompt_tokens.end() - params.n_ctx, prompt_tokens.end(), last_n_tokens.begin()); LOG_VERBOSE("input truncated", { - {"n_ctx", params.n_ctx}, - {"n_keep", params.n_keep}, - {"n_left", n_left}, - {"new_tokens", tokens_to_str(ctx, new_tokens.cbegin(), new_tokens.cend())}, - }); + { "n_ctx", params.n_ctx }, + { "n_keep", params.n_keep }, + { "n_left", n_left }, + { "new_tokens", tokens_to_str(ctx, new_tokens.cbegin(), new_tokens.cend()) }, + }); truncated = true; prompt_tokens = new_tokens; - } - else - { + } else { const size_t ps = num_prompt_tokens; std::fill(last_n_tokens.begin(), last_n_tokens.end() - ps, 0); std::copy(prompt_tokens.begin(), prompt_tokens.end(), last_n_tokens.end() - ps); @@ -298,35 +261,31 @@ struct llama_server_context // compare the evaluated prompt with the new prompt n_past = common_part(embd, prompt_tokens); embd = prompt_tokens; - if (n_past == num_prompt_tokens) - { + if (n_past == num_prompt_tokens) { // we have to evaluate at least 1 token to generate logits. n_past--; } LOG_VERBOSE("prompt ingested", { - {"n_past", n_past}, - {"cached", tokens_to_str(ctx, embd.cbegin(), embd.cbegin() + n_past)}, - {"to_eval", tokens_to_str(ctx, embd.cbegin() + n_past, embd.cend())}, - }); + { "n_past", n_past }, + { "cached", tokens_to_str(ctx, embd.cbegin(), embd.cbegin() + n_past) }, + { "to_eval", tokens_to_str(ctx, embd.cbegin() + n_past, embd.cend()) }, + }); has_next_token = true; } - void beginCompletion() - { + void beginCompletion() { // number of tokens to keep when resetting context n_remain = params.n_predict; llama_set_rng_seed(ctx, params.seed); } - completion_token_output nextToken() - { + completion_token_output nextToken() { completion_token_output result; result.tok = -1; - if (embd.size() >= (size_t)params.n_ctx) - { + if (embd.size() >= (size_t)params.n_ctx) { // Reset context const int n_left = (params.n_ctx - params.n_keep) / 2; @@ -336,36 +295,32 @@ struct llama_server_context n_past = params.n_keep; truncated = true; LOG_VERBOSE("input truncated", { - {"n_ctx", params.n_ctx}, - {"n_keep", params.n_keep}, - {"n_left", n_left}, - {"new_tokens", tokens_to_str(ctx, new_tokens.cbegin(), new_tokens.cend())}, - }); + { "n_ctx", params.n_ctx }, + { "n_keep", params.n_keep }, + { "n_left", n_left }, + { "new_tokens", tokens_to_str(ctx, new_tokens.cbegin(), new_tokens.cend()) }, + }); } - while (n_past < embd.size()) - { + while (n_past < embd.size()) { int n_eval = (int)embd.size() - n_past; - if (n_eval > params.n_batch) - { + if (n_eval > params.n_batch) { n_eval = params.n_batch; } - if (llama_eval(ctx, &embd[n_past], n_eval, n_past, params.n_threads)) - { + if (llama_eval(ctx, &embd[n_past], n_eval, n_past, params.n_threads)) { LOG_ERROR("failed to eval", { - {"n_eval", n_eval}, - {"n_past", n_past}, - {"n_threads", params.n_threads}, - {"embd", tokens_to_str(ctx, embd.cbegin() + n_past, embd.cend())}, - }); + { "n_eval", n_eval }, + { "n_past", n_past }, + { "n_threads", params.n_threads }, + { "embd", tokens_to_str(ctx, embd.cbegin() + n_past, embd.cend()) }, + }); has_next_token = false; return result; } n_past += n_eval; } - if (params.n_predict == 0) - { + if (params.n_predict == 0) { has_next_token = false; result.tok = llama_token_eos(); return result; @@ -388,64 +343,52 @@ struct llama_server_context const int32_t n_probs = params.n_probs; { - auto *logits = llama_get_logits(ctx); + auto * logits = llama_get_logits(ctx); auto n_vocab = llama_n_vocab(ctx); // Apply params.logit_bias map - for (const auto &it : params.logit_bias) - { + for (const auto & it : params.logit_bias) { logits[it.first] += it.second; } 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}); + 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}; + llama_token_data_array candidates_p = { candidates.data(), candidates.size(), false }; // Apply penalties float nl_logit = logits[llama_token_nl()]; auto last_n_repeat = std::min(std::min((int)last_n_tokens.size(), repeat_last_n), params.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); + 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) - { + last_n_tokens.data() + last_n_tokens.size() - last_n_repeat, + last_n_repeat, alpha_frequency, alpha_presence); + if (!penalize_nl) { logits[llama_token_nl()] = nl_logit; } - if (temp <= 0) - { + if (temp <= 0) { // Greedy sampling result.tok = llama_sample_token_greedy(ctx, &candidates_p); - if (n_probs > 0) - { + if (n_probs > 0) { llama_sample_softmax(ctx, &candidates_p); } - } - else - { - if (mirostat == 1) - { + } else { + if (mirostat == 1) { static float mirostat_mu = 2.0f * mirostat_tau; const int mirostat_m = 100; llama_sample_temperature(ctx, &candidates_p, temp); result.tok = llama_sample_token_mirostat(ctx, &candidates_p, mirostat_tau, mirostat_eta, mirostat_m, &mirostat_mu); - } - else if (mirostat == 2) - { + } else if (mirostat == 2) { static float mirostat_mu = 2.0f * mirostat_tau; llama_sample_temperature(ctx, &candidates_p, temp); result.tok = llama_sample_token_mirostat_v2(ctx, &candidates_p, mirostat_tau, mirostat_eta, &mirostat_mu); - } - else - { + } else { // Temperature sampling size_t min_keep = std::max(1, n_probs); llama_sample_top_k(ctx, &candidates_p, top_k, min_keep); @@ -457,8 +400,7 @@ struct llama_server_context } } - for (size_t i = 0; i < std::min(candidates_p.size, (size_t)n_probs); ++i) - { + for (size_t i = 0; i < std::min(candidates_p.size, (size_t) n_probs); ++i) { result.probs.push_back({candidates_p.data[i].id, candidates_p.data[i].p}); } last_n_tokens.erase(last_n_tokens.begin()); @@ -471,9 +413,8 @@ struct llama_server_context // decrement remaining sampling budget --n_remain; - if (!embd.empty() && embd.back() == llama_token_eos()) - { - // stopping_word = llama_token_to_str(ctx, embd.back()); + if (!embd.empty() && embd.back() == llama_token_eos()) { + //stopping_word = llama_token_to_str(ctx, embd.back()); has_next_token = false; stopped_eos = true; LOG_VERBOSE("eos token found", {}); @@ -484,28 +425,22 @@ struct llama_server_context return result; } - size_t findStoppingStrings(const std::string &text, const size_t last_token_size, - const stop_type type) - { + size_t findStoppingStrings(const std::string & text, const size_t last_token_size, + const stop_type type) { size_t stop_pos = std::string::npos; - for (const std::string &word : params.antiprompt) - { + for (const std::string & word : params.antiprompt) { size_t pos; - if (type == STOP_FULL) - { + if (type == STOP_FULL) { const size_t tmp = word.size() + last_token_size; const size_t from_pos = text.size() > tmp ? text.size() - tmp : 0; pos = text.find(word, from_pos); } - else - { + else { pos = find_partial_stop_string(word, text); } if (pos != std::string::npos && - (stop_pos == std::string::npos || pos < stop_pos)) - { - if (type == STOP_FULL) - { + (stop_pos == std::string::npos || pos < stop_pos)) { + if (type == STOP_FULL) { stopping_word = word; stopped_word = true; has_next_token = false; @@ -516,91 +451,74 @@ struct llama_server_context return stop_pos; } - completion_token_output doCompletion() - { + completion_token_output doCompletion() { const completion_token_output token_with_probs = nextToken(); const std::string token_text = token_with_probs.tok == -1 ? "" : llama_token_to_str(ctx, token_with_probs.tok); generated_text += token_text; - if (params.n_probs > 0) - { + if (params.n_probs > 0) { generated_token_probs.push_back(token_with_probs); } - if (multibyte_pending > 0) - { + if (multibyte_pending > 0) { multibyte_pending -= token_text.size(); - } - else if (token_text.size() == 1) - { + } else if (token_text.size() == 1) { const char c = token_text[0]; // 2-byte characters: 110xxxxx 10xxxxxx - if ((c & 0xE0) == 0xC0) - { + if ((c & 0xE0) == 0xC0) { multibyte_pending = 1; - // 3-byte characters: 1110xxxx 10xxxxxx 10xxxxxx - } - else if ((c & 0xF0) == 0xE0) - { + // 3-byte characters: 1110xxxx 10xxxxxx 10xxxxxx + } else if ((c & 0xF0) == 0xE0) { multibyte_pending = 2; - // 4-byte characters: 11110xxx 10xxxxxx 10xxxxxx 10xxxxxx - } - else if ((c & 0xF8) == 0xF0) - { + // 4-byte characters: 11110xxx 10xxxxxx 10xxxxxx 10xxxxxx + } else if ((c & 0xF8) == 0xF0) { multibyte_pending = 3; - } - else - { + } else { multibyte_pending = 0; } } - if (multibyte_pending > 0 && !has_next_token) - { + if (multibyte_pending > 0 && !has_next_token) { has_next_token = true; n_remain++; } - if (!has_next_token && n_remain == 0) - { + if (!has_next_token && n_remain == 0) { stopped_limit = true; } LOG_VERBOSE("next token", { - {"token", token_with_probs.tok}, - {"token_text", tokens_to_output_formatted_string(ctx, token_with_probs.tok)}, - {"has_next_token", has_next_token}, - {"n_remain", n_remain}, - {"num_tokens_predicted", num_tokens_predicted}, - {"stopped_eos", stopped_eos}, - {"stopped_word", stopped_word}, - {"stopped_limit", stopped_limit}, - {"stopping_word", stopping_word}, - }); + { "token", token_with_probs.tok }, + { "token_text", tokens_to_output_formatted_string(ctx, token_with_probs.tok) }, + { "has_next_token", has_next_token }, + { "n_remain", n_remain }, + { "num_tokens_predicted", num_tokens_predicted }, + { "stopped_eos", stopped_eos }, + { "stopped_word", stopped_word }, + { "stopped_limit", stopped_limit }, + { "stopping_word", stopping_word }, + }); return token_with_probs; } - std::vector getEmbedding() - { + std::vector getEmbedding() { static const int n_embd = llama_n_embd(ctx); - if (!params.embedding) - { + if (!params.embedding) { LOG_WARNING("embedding disabled", { - {"params.embedding", params.embedding}, - }); + { "params.embedding", params.embedding }, + }); return std::vector(n_embd, 0.0f); } - const float *data = llama_get_embeddings(ctx); + const float * data = llama_get_embeddings(ctx); std::vector embedding(data, data + n_embd); return embedding; } }; -static void server_print_usage(const char *argv0, const gpt_params ¶ms, - const server_params &sparams) -{ +static void server_print_usage(const char * argv0, const gpt_params & params, + const server_params & sparams) { fprintf(stderr, "usage: %s [options]\n", argv0); fprintf(stderr, "\n"); fprintf(stderr, "options:\n"); @@ -611,12 +529,10 @@ static void server_print_usage(const char *argv0, const gpt_params ¶ms, fprintf(stderr, " -b N, --batch-size N batch size for prompt processing (default: %d)\n", params.n_batch); fprintf(stderr, " --memory-f32 use f32 instead of f16 for memory key+value (default: disabled)\n"); fprintf(stderr, " not recommended: doubles context memory required and no measurable increase in quality\n"); - if (llama_mlock_supported()) - { + if (llama_mlock_supported()) { fprintf(stderr, " --mlock force system to keep model in RAM rather than swapping or compressing\n"); } - if (llama_mmap_supported()) - { + if (llama_mmap_supported()) { fprintf(stderr, " --no-mmap do not memory-map model (slower load but may reduce pageouts if not using mlock)\n"); } #ifdef LLAMA_SUPPORTS_GPU_OFFLOAD @@ -642,113 +558,78 @@ static void server_print_usage(const char *argv0, const gpt_params ¶ms, fprintf(stderr, "\n"); } -static void server_params_parse(int argc, char **argv, server_params &sparams, - gpt_params ¶ms) -{ +static void server_params_parse(int argc, char ** argv, server_params & sparams, + gpt_params & params) { gpt_params default_params; server_params default_sparams; std::string arg; bool invalid_param = false; - for (int i = 1; i < argc; i++) - { + for (int i = 1; i < argc; i++) { arg = argv[i]; - if (arg == "--port") - { - if (++i >= argc) - { + if (arg == "--port") { + if (++i >= argc) { invalid_param = true; break; } sparams.port = std::stoi(argv[i]); - } - else if (arg == "--host") - { - if (++i >= argc) - { + } else if (arg == "--host") { + if (++i >= argc) { invalid_param = true; break; } sparams.hostname = argv[i]; - } - else if (arg == "--path") - { - if (++i >= argc) - { + } else if (arg == "--path") { + if (++i >= argc) { invalid_param = true; break; } sparams.public_path = argv[i]; - } - else if (arg == "--timeout" || arg == "-to") - { - if (++i >= argc) - { + } else if (arg == "--timeout" || arg == "-to") { + if (++i >= argc) { invalid_param = true; break; } sparams.read_timeout = std::stoi(argv[i]); sparams.write_timeout = std::stoi(argv[i]); - } - else if (arg == "-m" || arg == "--model") - { - if (++i >= argc) - { + } else if (arg == "-m" || arg == "--model") { + if (++i >= argc) { invalid_param = true; break; } params.model = argv[i]; - } - else if (arg == "-a" || arg == "--alias") - { - if (++i >= argc) - { + } else if (arg == "-a" || arg == "--alias") { + if (++i >= argc) { invalid_param = true; break; } params.model_alias = argv[i]; - } - else if (arg == "-h" || arg == "--help") - { + } else if (arg == "-h" || arg == "--help") { server_print_usage(argv[0], default_params, default_sparams); exit(0); - } - else if (arg == "-c" || arg == "--ctx-size" || arg == "--ctx_size") - { - if (++i >= argc) - { + } else if (arg == "-c" || arg == "--ctx-size" || arg == "--ctx_size") { + if (++i >= argc) { invalid_param = true; break; } params.n_ctx = std::stoi(argv[i]); - } - else if (arg == "--memory-f32" || arg == "--memory_f32") - { + } else if (arg == "--memory-f32" || arg == "--memory_f32") { params.memory_f16 = false; - } - else if (arg == "--threads" || arg == "-t") - { - if (++i >= argc) - { + } else if (arg == "--threads" || arg == "-t") { + if (++i >= argc) { invalid_param = true; break; } params.n_threads = std::stoi(argv[i]); - } - else if (arg == "-b" || arg == "--batch-size") - { - if (++i >= argc) - { + } else if (arg == "-b" || arg == "--batch-size") { + if (++i >= argc) { invalid_param = true; break; } params.n_batch = std::stoi(argv[i]); params.n_batch = std::min(512, params.n_batch); - } - else if (arg == "--gpu-layers" || arg == "-ngl" || arg == "--n-gpu-layers") - { - if (++i >= argc) - { + } else if (arg == "--gpu-layers" || arg == "-ngl" || arg == "--n-gpu-layers") { + if (++i >= argc) { invalid_param = true; break; } @@ -757,13 +638,11 @@ static void server_params_parse(int argc, char **argv, server_params &sparams, #else LOG_WARNING("Not compiled with GPU offload support, --n-gpu-layers option will be ignored. " "See main README.md for information on enabling GPU BLAS support", - {{"n_gpu_layers", params.n_gpu_layers}}); + {{ "n_gpu_layers", params.n_gpu_layers }}); #endif } - else if (arg == "--tensor-split" || arg == "-ts") - { - if (++i >= argc) - { + else if (arg == "--tensor-split" || arg == "-ts") { + if (++i >= argc) { invalid_param = true; break; } @@ -771,38 +650,29 @@ static void server_params_parse(int argc, char **argv, server_params &sparams, std::string arg_next = argv[i]; // split string by , and / - const std::regex regex{R"([,/]+)"}; - std::sregex_token_iterator it{arg_next.begin(), arg_next.end(), regex, -1}; - std::vector split_arg{it, {}}; + const std::regex regex{ R"([,/]+)" }; + std::sregex_token_iterator it{ arg_next.begin(), arg_next.end(), regex, -1 }; + std::vector split_arg{ it, {} }; GGML_ASSERT(split_arg.size() <= LLAMA_MAX_DEVICES); - for (size_t i_device = 0; i_device < LLAMA_MAX_DEVICES; ++i_device) - { - if (i_device < split_arg.size()) - { + for (size_t i_device = 0; i_device < LLAMA_MAX_DEVICES; ++i_device) { + if (i_device < split_arg.size()) { params.tensor_split[i_device] = std::stof(split_arg[i_device]); - } - else - { + } else { params.tensor_split[i_device] = 0.0f; } } #else LOG_WARNING("llama.cpp was compiled without cuBLAS. It is not possible to set a tensor split.", {}); #endif // GGML_USE_CUBLAS - } - else if (arg == "--low-vram" || arg == "-lv") - { + } else if (arg == "--low-vram" || arg == "-lv") { #ifdef GGML_USE_CUBLAS params.low_vram = true; #else fprintf(stderr, "warning: llama.cpp was compiled without cuBLAS. It is not possible to set lower vram usage.\n"); #endif // GGML_USE_CUBLAS - } - else if (arg == "--main-gpu" || arg == "-mg") - { - if (++i >= argc) - { + }else if (arg == "--main-gpu" || arg == "-mg") { + if (++i >= argc) { invalid_param = true; break; } @@ -811,173 +681,146 @@ static void server_params_parse(int argc, char **argv, server_params &sparams, #else LOG_WARNING("llama.cpp was compiled without cuBLAS. It is not possible to set a main GPU.", {}); #endif - } - else if (arg == "--lora") - { - if (++i >= argc) - { + } else if (arg == "--lora") { + if (++i >= argc) { invalid_param = true; break; } params.lora_adapter = argv[i]; params.use_mmap = false; - } - else if (arg == "--lora-base") - { - if (++i >= argc) - { + } else if (arg == "--lora-base") { + if (++i >= argc) { invalid_param = true; break; } params.lora_base = argv[i]; - } - else if (arg == "-v" || arg == "--verbose") - { + } else if (arg == "-v" || arg == "--verbose") { #if SERVER_VERBOSE != 1 LOG_WARNING("server.cpp is not built with verbose logging.", {}); #else server_verbose = true; #endif - } - else if (arg == "--mlock") - { + } else if (arg == "--mlock") { params.use_mlock = true; - } - else if (arg == "--no-mmap") - { + } else if (arg == "--no-mmap") { params.use_mmap = false; - } - else if (arg == "--embedding") - { + } else if (arg == "--embedding") { params.embedding = true; - } - else - { + } else { fprintf(stderr, "error: unknown argument: %s\n", arg.c_str()); server_print_usage(argv[0], default_params, default_sparams); exit(1); } } - if (invalid_param) - { + if (invalid_param) { fprintf(stderr, "error: invalid parameter for argument: %s\n", arg.c_str()); server_print_usage(argv[0], default_params, default_sparams); exit(1); } } -static json format_generation_settings(llama_server_context &llama) -{ +static json format_generation_settings(llama_server_context & llama) { const auto eos_bias = llama.params.logit_bias.find(llama_token_eos()); const bool ignore_eos = eos_bias != llama.params.logit_bias.end() && - eos_bias->second < 0.0f && std::isinf(eos_bias->second); + eos_bias->second < 0.0f && std::isinf(eos_bias->second); - return json{ - {"n_ctx", llama.params.n_ctx}, - {"model", llama.params.model_alias}, - {"seed", llama.params.seed}, - {"temp", llama.params.temp}, - {"top_k", llama.params.top_k}, - {"top_p", llama.params.top_p}, - {"tfs_z", llama.params.tfs_z}, - {"typical_p", llama.params.typical_p}, - {"repeat_last_n", llama.params.repeat_last_n}, - {"repeat_penalty", llama.params.repeat_penalty}, - {"presence_penalty", llama.params.presence_penalty}, - {"frequency_penalty", llama.params.frequency_penalty}, - {"mirostat", llama.params.mirostat}, - {"mirostat_tau", llama.params.mirostat_tau}, - {"mirostat_eta", llama.params.mirostat_eta}, - {"penalize_nl", llama.params.penalize_nl}, - {"stop", llama.params.antiprompt}, - {"n_predict", llama.params.n_predict}, - {"n_keep", llama.params.n_keep}, - {"ignore_eos", ignore_eos}, - {"stream", llama.stream}, - {"logit_bias", llama.params.logit_bias}, - {"n_probs", llama.params.n_probs}, + return json { + { "seed", llama.params.seed }, + { "temp", llama.params.temp }, + { "top_k", llama.params.top_k }, + { "top_p", llama.params.top_p }, + { "tfs_z", llama.params.tfs_z }, + { "typical_p", llama.params.typical_p }, + { "repeat_last_n", llama.params.repeat_last_n }, + { "repeat_penalty", llama.params.repeat_penalty }, + { "presence_penalty", llama.params.presence_penalty }, + { "frequency_penalty", llama.params.frequency_penalty }, + { "mirostat", llama.params.mirostat }, + { "mirostat_tau", llama.params.mirostat_tau }, + { "mirostat_eta", llama.params.mirostat_eta }, + { "penalize_nl", llama.params.penalize_nl }, + { "stop", llama.params.antiprompt }, + { "n_predict", llama.params.n_predict }, + { "n_keep", llama.params.n_keep }, + { "ignore_eos", ignore_eos }, + { "stream", llama.stream }, + { "logit_bias", llama.params.logit_bias }, + { "n_probs", llama.params.n_probs }, }; } -static json format_embedding_response(llama_server_context &llama) -{ - return json{ - {"embedding", llama.getEmbedding()}, +static json format_embedding_response(llama_server_context & llama) { + return json { + { "embedding", llama.getEmbedding() }, }; } -static json format_timings(llama_server_context &llama) -{ +static json format_timings(llama_server_context & llama) { const auto timings = llama_get_timings(llama.ctx); assert(timings.n_eval == llama.num_tokens_predicted); - return json{ - {"prompt_n", timings.n_eval}, - {"prompt_ms", timings.t_p_eval_ms}, - {"prompt_per_token_ms", timings.t_p_eval_ms / timings.n_p_eval}, - {"prompt_per_second", 1e3 / timings.t_p_eval_ms * timings.n_p_eval}, + return json { + { "prompt_n", timings.n_eval }, + { "prompt_ms", timings.t_p_eval_ms }, + { "prompt_per_token_ms", timings.t_p_eval_ms / timings.n_p_eval }, + { "prompt_per_second", 1e3 / timings.t_p_eval_ms * timings.n_p_eval }, - {"predicted_n", timings.n_eval}, - {"predicted_ms", timings.t_eval_ms}, - {"predicted_per_token_ms", timings.t_eval_ms / timings.n_eval}, - {"predicted_per_second", 1e3 / timings.t_eval_ms * timings.n_eval}, + { "predicted_n", timings.n_eval }, + { "predicted_ms", timings.t_eval_ms }, + { "predicted_per_token_ms", timings.t_eval_ms / timings.n_eval }, + { "predicted_per_second", 1e3 / timings.t_eval_ms * timings.n_eval }, }; } -static json format_final_response(llama_server_context &llama, const std::string &content, const std::vector &probs) -{ +static json format_final_response(llama_server_context & llama, const std::string & content, const std::vector & probs) { - json res = json{ - {"content", content}, - {"stop", true}, - {"model", llama.params.model_alias}, - {"tokens_predicted", llama.num_tokens_predicted}, - {"tokens_evaluated", llama.num_prompt_tokens}, - {"generation_settings", format_generation_settings(llama)}, - {"prompt", llama.params.prompt}, - {"truncated", llama.truncated}, - {"stopped_eos", llama.stopped_eos}, - {"stopped_word", llama.stopped_word}, - {"stopped_limit", llama.stopped_limit}, - {"stopping_word", llama.stopping_word}, - {"tokens_cached", llama.n_past}, - {"tokens_predicted", llama.num_tokens_predicted}, - {"timings", format_timings(llama)}, + json res = json { + { "content", content }, + { "stop", true }, + { "model", llama.params.model_alias }, + { "tokens_predicted", llama.num_tokens_predicted }, + { "tokens_evaluated", llama.num_prompt_tokens }, + { "generation_settings", format_generation_settings(llama) }, + { "prompt", llama.params.prompt }, + { "truncated", llama.truncated }, + { "stopped_eos", llama.stopped_eos }, + { "stopped_word", llama.stopped_word }, + { "stopped_limit", llama.stopped_limit }, + { "stopping_word", llama.stopping_word }, + { "tokens_cached", llama.n_past }, + { "tokens_predicted", llama.num_tokens_predicted }, + { "timings", format_timings(llama) }, }; - if (llama.params.n_probs > 0) - { + if (llama.params.n_probs > 0) { res["completion_probabilities"] = probs_vector_to_json(llama.ctx, probs); } return res; } -static json format_partial_response(llama_server_context &llama, const std::string &content, const std::vector &probs) -{ - json res = json{ - {"content", content}, - {"stop", false}, +static json format_partial_response(llama_server_context & llama, const std::string & content, const std::vector & probs) { + json res = json { + { "content", content }, + { "stop", false }, }; - if (llama.params.n_probs > 0) - { + if (llama.params.n_probs > 0) { res["completion_probabilities"] = probs_vector_to_json(llama.ctx, probs); } return res; } -static json format_tokenizer_response(const std::vector &tokens) -{ - return json{ - {"tokens", tokens}}; +static json format_tokenizer_response(const std::vector & tokens) { + return json { + { "tokens", tokens } + }; } -static void parse_options_completion(const json &body, llama_server_context &llama) -{ +static void parse_options_completion(const json & body, llama_server_context & llama) { gpt_params default_params; llama.stream = body.value("stream", false); @@ -1001,28 +844,20 @@ static void parse_options_completion(const json &body, llama_server_context &lla llama.params.n_probs = body.value("n_probs", default_params.n_probs); llama.params.logit_bias.clear(); - if (body.value("ignore_eos", false)) - { + if (body.value("ignore_eos", false)) { llama.params.logit_bias[llama_token_eos()] = -INFINITY; } - const auto &logit_bias = body.find("logit_bias"); - if (logit_bias != body.end() && logit_bias->is_array()) - { + const auto & logit_bias = body.find("logit_bias"); + if (logit_bias != body.end() && logit_bias->is_array()) { const int n_vocab = llama_n_vocab(llama.ctx); - for (const auto &el : *logit_bias) - { - if (el.is_array() && el.size() == 2 && el[0].is_number_integer()) - { + for (const auto & el : *logit_bias) { + if (el.is_array() && el.size() == 2 && el[0].is_number_integer()) { llama_token tok = el[0].get(); - if (tok >= 0 && tok < n_vocab) - { - if (el[1].is_number()) - { + if (tok >= 0 && tok < n_vocab) { + if (el[1].is_number()) { llama.params.logit_bias[tok] = el[1].get(); - } - else if (el[1].is_boolean() && !el[1].get()) - { + } else if (el[1].is_boolean() && !el[1].get()) { llama.params.logit_bias[tok] = -INFINITY; } } @@ -1031,13 +866,10 @@ static void parse_options_completion(const json &body, llama_server_context &lla } llama.params.antiprompt.clear(); - const auto &stop = body.find("stop"); - if (stop != body.end() && stop->is_array()) - { - for (const auto &word : *stop) - { - if (!word.empty()) - { + const auto & stop = body.find("stop"); + if (stop != body.end() && stop->is_array()) { + for (const auto & word : *stop) { + if (!word.empty()) { llama.params.antiprompt.push_back(word); } } @@ -1046,25 +878,23 @@ static void parse_options_completion(const json &body, llama_server_context &lla LOG_VERBOSE("completion parameters parsed", format_generation_settings(llama)); } -static void log_server_request(const Request &req, const Response &res) -{ +static void log_server_request(const Request & req, const Response & res) { LOG_INFO("request", { - {"remote_addr", req.remote_addr}, - {"remote_port", req.remote_port}, - {"status", res.status}, - {"method", req.method}, - {"path", req.path}, - {"params", req.params}, - }); + { "remote_addr", req.remote_addr }, + { "remote_port", req.remote_port }, + { "status", res.status }, + { "method", req.method }, + { "path", req.path }, + { "params", req.params }, + }); LOG_VERBOSE("request", { - {"request", req.body}, - {"response", res.body}, - }); + { "request", req.body }, + { "response", res.body }, + }); } -int main(int argc, char **argv) -{ +int main(int argc, char ** argv) { // own arguments required by this example gpt_params params; server_params sparams; @@ -1074,53 +904,54 @@ int main(int argc, char **argv) server_params_parse(argc, argv, sparams, params); - if (params.model_alias == "unknown") - { + if (params.model_alias == "unknown") { params.model_alias = params.model; } llama_backend_init(params.numa); - LOG_INFO("build info", {{"build", BUILD_NUMBER}, - {"commit", BUILD_COMMIT}}); + LOG_INFO("build info", { + { "build", BUILD_NUMBER }, + { "commit", BUILD_COMMIT }, + }); LOG_INFO("system info", { - {"n_threads", params.n_threads}, - {"total_threads", std::thread::hardware_concurrency()}, - {"system_info", llama_print_system_info()}, - }); + { "n_threads", params.n_threads }, + { "total_threads", std::thread::hardware_concurrency() }, + { "system_info", llama_print_system_info() }, + }); // load the model - if (!llama.loadModel(params)) - { + if (!llama.loadModel(params)) { return 1; } Server svr; - svr.set_default_headers({{"Server", "llama.cpp"}, - {"Access-Control-Allow-Origin", "*"}, - {"Access-Control-Allow-Headers", "content-type"}}); + svr.set_default_headers({ + { "Server", "llama.cpp" }, + { "Access-Control-Allow-Origin", "*" }, + { "Access-Control-Allow-Headers", "content-type" }, + }); // this is only called if no index.html is found in the public --path - svr.Get("/", [](const Request &, Response &res) - { - res.set_content(reinterpret_cast(&index_html), index_html_len, "text/html"); - return false; }); + svr.Get("/", [](const Request &, Response & res) { + res.set_content(reinterpret_cast(&index_html), index_html_len, "text/html"); + return false; + }); // this is only called if no index.js is found in the public --path - svr.Get("/index.js", [](const Request &, Response &res) - { + svr.Get("/index.js", [](const Request &, Response & res) { res.set_content(reinterpret_cast(&index_js), index_js_len, "text/javascript"); - return false; }); + return false; + }); // this is only called if no index.html is found in the public --path - svr.Get("/completion.js", [](const Request &, Response &res) - { - res.set_content(reinterpret_cast(&completion_js), completion_js_len, "application/javascript"); - return false; }); + svr.Get("/completion.js", [](const Request &, Response & res) { + res.set_content(reinterpret_cast(&completion_js), completion_js_len, "application/javascript"); + return false; + }); - svr.Post("/completion", [&llama](const Request &req, Response &res) - { + svr.Post("/completion", [&llama](const Request & req, Response & res) { auto lock = llama.lock(); llama.rewind(); @@ -1225,28 +1056,29 @@ int main(int argc, char **argv) return true; }; res.set_chunked_content_provider("text/event-stream", chunked_content_provider); - } }); + } + }); - svr.Get("/model.json", [&llama](const Request &, Response &res) - { + svr.Get("/model.json", [&llama](const Request &, Response & res) { const json data = format_generation_settings(llama); - return res.set_content(data.dump(), "application/json"); }); + return res.set_content(data.dump(), "application/json"); + }); - svr.Options(R"(/.*)", [](const Request &, Response &res) - { return res.set_content("", "application/json"); }); + svr.Options(R"(/.*)", [](const Request &, Response & res) { + return res.set_content("", "application/json"); + }); - svr.Post("/tokenize", [&llama](const Request &req, Response &res) - { + svr.Post("/tokenize", [&llama](const Request & req, Response & res) { auto lock = llama.lock(); const json body = json::parse(req.body); const std::string content = body.value("content", ""); const std::vector tokens = llama_tokenize(llama.ctx, content, false); const json data = format_tokenizer_response(tokens); - return res.set_content(data.dump(), "application/json"); }); + return res.set_content(data.dump(), "application/json"); + }); - svr.Post("/embedding", [&llama](const Request &req, Response &res) - { + svr.Post("/embedding", [&llama](const Request & req, Response & res) { auto lock = llama.lock(); const json body = json::parse(req.body); @@ -1260,12 +1092,12 @@ int main(int argc, char **argv) llama.doCompletion(); const json data = format_embedding_response(llama); - return res.set_content(data.dump(), "application/json"); }); + return res.set_content(data.dump(), "application/json"); + }); svr.set_logger(log_server_request); - svr.set_exception_handler([](const Request &, Response &res, std::exception_ptr ep) - { + svr.set_exception_handler([](const Request &, Response & res, std::exception_ptr ep) { const auto * fmt = "500 Internal Server Error\n%s"; char buf[BUFSIZ]; try { @@ -1276,19 +1108,20 @@ int main(int argc, char **argv) snprintf(buf, sizeof(buf), fmt, "Unknown Exception"); } res.set_content(buf, "text/plain"); - res.status = 500; }); + res.status = 500; + }); - svr.set_error_handler([](const Request &, Response &res) - { + svr.set_error_handler([](const Request &, Response & res) { res.set_content("File Not Found", "text/plain"); - res.status = 404; }); + res.status = 404; + }); + // set timeouts and change hostname and port svr.set_read_timeout(sparams.read_timeout); svr.set_write_timeout(sparams.write_timeout); - if (!svr.bind_to_port(sparams.hostname, sparams.port)) - { + if (!svr.bind_to_port(sparams.hostname, sparams.port)) { fprintf(stderr, "\ncouldn't bind to server socket: hostname=%s port=%d\n\n", sparams.hostname.c_str(), sparams.port); return 1; } @@ -1300,12 +1133,11 @@ int main(int argc, char **argv) fprintf(stdout, "\nllama server listening at http://%s:%d\n\n", sparams.hostname.c_str(), sparams.port); LOG_INFO("HTTP server listening", { - {"hostname", sparams.hostname}, - {"port", sparams.port}, - }); + { "hostname", sparams.hostname }, + { "port", sparams.port }, + }); - if (!svr.listen_after_bind()) - { + if (!svr.listen_after_bind()) { return 1; }