@fugood/llama.node 0.3.0 → 0.3.2

package/src/llama.cpp/common/common.cpp

@@ -3,6 +3,7 @@
 #endif
 
 #include "common.h"
+#include "log.h"
 // Change JSON_ASSERT from assert() to GGML_ASSERT:
 #define JSON_ASSERT GGML_ASSERT
 #include "json.hpp"
@@ -25,6 +26,7 @@
 #include <unordered_map>
 #include <unordered_set>
 #include <vector>
+#include <thread>
 
 #if defined(__APPLE__) && defined(__MACH__)
 #include <sys/types.h>
@@ -48,7 +50,6 @@
 #if defined(LLAMA_USE_CURL)
 #include <curl/curl.h>
 #include <curl/easy.h>
-#include <thread>
 #include <future>
 #endif
 
@@ -56,14 +57,6 @@
 #pragma warning(disable: 4244 4267) // possible loss of data
 #endif
 
-#if (defined(GGML_USE_CUDA) || defined(GGML_USE_SYCL))
-#define GGML_USE_CUDA_SYCL
-#endif
-
-#if (defined(GGML_USE_CUDA) || defined(GGML_USE_SYCL)) || defined(GGML_USE_VULKAN)
-#define GGML_USE_CUDA_SYCL_VULKAN
-#endif
-
 #if defined(LLAMA_USE_CURL)
 #ifdef __linux__
 #include <linux/limits.h>
@@ -110,8 +103,34 @@ int32_t cpu_get_num_physical_cores() {
     if (result == 0) {
         return num_physical_cores;
     }
-#elif defined(_WIN32)
-    //TODO: Implement
+#elif defined(_WIN32) && (_WIN32_WINNT >= 0x0601) && !defined(__MINGW64__) // windows 7 and later
+    // TODO: windows + arm64 + mingw64
+    unsigned int n_threads_win = std::thread::hardware_concurrency();
+    unsigned int default_threads = n_threads_win > 0 ? (n_threads_win <= 4 ? n_threads_win : n_threads_win / 2) : 4;
+
+    DWORD buffer_size = 0;
+    if (!GetLogicalProcessorInformationEx(RelationProcessorCore, nullptr, &buffer_size)) {
+        if (GetLastError() != ERROR_INSUFFICIENT_BUFFER) {
+            return default_threads;
+        }
+    }
+
+    std::vector<char> buffer(buffer_size);
+    if (!GetLogicalProcessorInformationEx(RelationProcessorCore, reinterpret_cast<PSYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX>(buffer.data()), &buffer_size)) {
+        return default_threads;
+    }
+
+    int32_t num_physical_cores = 0;
+    PSYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX info = reinterpret_cast<PSYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX>(buffer.data());
+    while (buffer_size > 0) {
+        if (info->Relationship == RelationProcessorCore) {
+            num_physical_cores += info->Processor.GroupCount;
+        }
+        buffer_size -= info->Size;
+        info = reinterpret_cast<PSYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX>(reinterpret_cast<char*>(info) + info->Size);
+    }
+
+    return num_physical_cores > 0 ? num_physical_cores : default_threads;
 #endif
     unsigned int n_threads = std::thread::hardware_concurrency();
     return n_threads > 0 ? (n_threads <= 4 ? n_threads : n_threads / 2) : 4;
@@ -156,1567 +175,223 @@ static int cpu_count_math_cpus(int n_cpu) {
         if (pin_cpu(cpu)) {
             return -1;
         }
-        if (is_running_on_efficiency_core()) {
-            continue; // efficiency cores harm lockstep threading
-        }
-        ++cpu; // hyperthreading isn't useful for linear algebra
-        ++result;
-    }
-    return result;
-}
-
-#endif // __x86_64__ && __linux__
-
-/**
- * Returns number of CPUs on system that are useful for math.
- */
-int32_t cpu_get_num_math() {
-#if defined(__x86_64__) && defined(__linux__) && !defined(__ANDROID__)
-    int n_cpu = sysconf(_SC_NPROCESSORS_ONLN);
-    if (n_cpu < 1) {
-        return cpu_get_num_physical_cores();
-    }
-    if (is_hybrid_cpu()) {
-        cpu_set_t affinity;
-        if (!pthread_getaffinity_np(pthread_self(), sizeof(affinity), &affinity)) {
-            int result = cpu_count_math_cpus(n_cpu);
-            pthread_setaffinity_np(pthread_self(), sizeof(affinity), &affinity);
-            if (result > 0) {
-                return result;
-            }
-        }
-    }
-#endif
-    return cpu_get_num_physical_cores();
-}
-
-//
-// CLI argument parsing
-//
-
-void gpt_params_handle_hf_token(gpt_params & params) {
-    if (params.hf_token.empty() && std::getenv("HF_TOKEN")) {
-        params.hf_token = std::getenv("HF_TOKEN");
-    }
-}
-
-void gpt_params_handle_model_default(gpt_params & params) {
-    if (!params.hf_repo.empty()) {
-        // short-hand to avoid specifying --hf-file -> default it to --model
-        if (params.hf_file.empty()) {
-            if (params.model.empty()) {
-                throw std::invalid_argument("error: --hf-repo requires either --hf-file or --model\n");
-            }
-            params.hf_file = params.model;
-        } else if (params.model.empty()) {
-            params.model = fs_get_cache_file(string_split(params.hf_file, '/').back());
-        }
-    } else if (!params.model_url.empty()) {
-        if (params.model.empty()) {
-            auto f = string_split(params.model_url, '#').front();
-            f = string_split(f, '?').front();
-            params.model = fs_get_cache_file(string_split(f, '/').back());
-        }
-    } else if (params.model.empty()) {
-        params.model = DEFAULT_MODEL_PATH;
-    }
-}
-
-bool gpt_params_parse_ex(int argc, char ** argv, gpt_params & params) {
-    bool invalid_param = false;
-    std::string arg;
-    const std::string arg_prefix = "--";
-    llama_sampling_params & sparams = params.sparams;
-
-    for (int i = 1; i < argc; i++) {
-        arg = argv[i];
-        if (arg.compare(0, arg_prefix.size(), arg_prefix) == 0) {
-            std::replace(arg.begin(), arg.end(), '_', '-');
-        }
-        if (!gpt_params_find_arg(argc, argv, arg, params, i, invalid_param)) {
-            throw std::invalid_argument("error: unknown argument: " + arg);
-        }
-        if (invalid_param) {
-            throw std::invalid_argument("error: invalid parameter for argument: " + arg);
-        }
-    }
-
-    if (params.prompt_cache_all && (params.interactive || params.interactive_first)) {
-        throw std::invalid_argument("error: --prompt-cache-all not supported in interactive mode yet\n");
-    }
-
-    gpt_params_handle_model_default(params);
-
-    gpt_params_handle_hf_token(params);
-
-    if (params.escape) {
-        string_process_escapes(params.prompt);
-        string_process_escapes(params.input_prefix);
-        string_process_escapes(params.input_suffix);
-        string_process_escapes(sparams.cfg_negative_prompt);
-        for (auto & antiprompt : params.antiprompt) {
-            string_process_escapes(antiprompt);
-        }
-    }
-
-    if (!params.kv_overrides.empty()) {
-        params.kv_overrides.emplace_back();
-        params.kv_overrides.back().key[0] = 0;
-    }
-
-    return true;
-}
-
-bool gpt_params_parse(int argc, char ** argv, gpt_params & params) {
-    const auto params_org = params; // the example can modify the default params
-
-    try {
-        if (!gpt_params_parse_ex(argc, argv, params) || params.usage) {
-            params = params_org;
-            params.usage = true;
-            return false;
-        }
-    } catch (const std::invalid_argument & ex) {
-        fprintf(stderr, "%s\n", ex.what());
-        params = params_org;
-        return false;
-    }
-
-    return true;
-}
-
-#define CHECK_ARG if (++i >= argc) { invalid_param = true; return true; }
-
-bool gpt_params_find_arg(int argc, char ** argv, const std::string & arg, gpt_params & params, int & i, bool & invalid_param) {
-    const char split_delim = ',';
-
-    llama_sampling_params & sparams = params.sparams;
-
-    if (arg == "-s" || arg == "--seed") {
-        CHECK_ARG
-        // TODO: this is temporary, in the future the sampling state will be moved fully to llama_sampling_context.
-        params.seed = std::stoul(argv[i]);
-        sparams.seed = std::stoul(argv[i]);
-        return true;
-    }
-    if (arg == "-t" || arg == "--threads") {
-        CHECK_ARG
-        params.n_threads = std::stoi(argv[i]);
-        if (params.n_threads <= 0) {
-            params.n_threads = std::thread::hardware_concurrency();
-        }
-        return true;
-    }
-    if (arg == "-tb" || arg == "--threads-batch") {
-        CHECK_ARG
-        params.n_threads_batch = std::stoi(argv[i]);
-        if (params.n_threads_batch <= 0) {
-            params.n_threads_batch = std::thread::hardware_concurrency();
-        }
-        return true;
-    }
-    if (arg == "-td" || arg == "--threads-draft") {
-        CHECK_ARG
-        params.n_threads_draft = std::stoi(argv[i]);
-        if (params.n_threads_draft <= 0) {
-            params.n_threads_draft = std::thread::hardware_concurrency();
-        }
-        return true;
-    }
-    if (arg == "-tbd" || arg == "--threads-batch-draft") {
-        CHECK_ARG
-        params.n_threads_batch_draft = std::stoi(argv[i]);
-        if (params.n_threads_batch_draft <= 0) {
-            params.n_threads_batch_draft = std::thread::hardware_concurrency();
-        }
-        return true;
-    }
-    if (arg == "-p" || arg == "--prompt") {
-        CHECK_ARG
-        params.prompt = argv[i];
-        return true;
-    }
-    if (arg == "-e" || arg == "--escape") {
-        params.escape = true;
-        return true;
-    }
-    if (arg == "--no-escape") {
-        params.escape = false;
-        return true;
-    }
-    if (arg == "--prompt-cache") {
-        CHECK_ARG
-        params.path_prompt_cache = argv[i];
-        return true;
-    }
-    if (arg == "--prompt-cache-all") {
-        params.prompt_cache_all = true;
-        return true;
-    }
-    if (arg == "--prompt-cache-ro") {
-        params.prompt_cache_ro = true;
-        return true;
-    }
-    if (arg == "-bf" || arg == "--binary-file") {
-        CHECK_ARG
-        std::ifstream file(argv[i], std::ios::binary);
-        if (!file) {
-            fprintf(stderr, "error: failed to open file '%s'\n", argv[i]);
-            invalid_param = true;
-            return true;
-        }
-        // store the external file name in params
-        params.prompt_file = argv[i];
-        std::ostringstream ss;
-        ss << file.rdbuf();
-        params.prompt = ss.str();
-        fprintf(stderr, "Read %zu bytes from binary file %s\n", params.prompt.size(), argv[i]);
-        return true;
-    }
-    if (arg == "-f" || arg == "--file") {
-        CHECK_ARG
-        std::ifstream file(argv[i]);
-        if (!file) {
-            fprintf(stderr, "error: failed to open file '%s'\n", argv[i]);
-            invalid_param = true;
-            return true;
-        }
-        // store the external file name in params
-        params.prompt_file = argv[i];
-        std::copy(std::istreambuf_iterator<char>(file), std::istreambuf_iterator<char>(), back_inserter(params.prompt));
-        if (!params.prompt.empty() && params.prompt.back() == '\n') {
-            params.prompt.pop_back();
-        }
-        return true;
-    }
-    if (arg == "--in-file") {
-        CHECK_ARG
-        std::ifstream file(argv[i]);
-        if (!file) {
-            fprintf(stderr, "error: failed to open file '%s'\n", argv[i]);
-            invalid_param = true;
-            return true;
-        }
-        params.in_files.push_back(argv[i]);
-        return true;
-    }
-    if (arg == "-n" || arg == "--predict" || arg == "--n-predict") {
-        CHECK_ARG
-        params.n_predict = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "--top-k") {
-        CHECK_ARG
-        sparams.top_k = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "-c" || arg == "--ctx-size") {
-        CHECK_ARG
-        params.n_ctx = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "--grp-attn-n" || arg == "-gan") {
-        CHECK_ARG
-        params.grp_attn_n = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "--grp-attn-w" || arg == "-gaw") {
-        CHECK_ARG
-        params.grp_attn_w = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "--rope-freq-base") {
-        CHECK_ARG
-        params.rope_freq_base = std::stof(argv[i]);
-        return true;
-    }
-    if (arg == "--rope-freq-scale") {
-        CHECK_ARG
-        params.rope_freq_scale = std::stof(argv[i]);
-        return true;
-    }
-    if (arg == "--rope-scaling") {
-        CHECK_ARG
-        std::string value(argv[i]);
-        /**/ if (value == "none") { params.rope_scaling_type = LLAMA_ROPE_SCALING_TYPE_NONE; }
-        else if (value == "linear") { params.rope_scaling_type = LLAMA_ROPE_SCALING_TYPE_LINEAR; }
-        else if (value == "yarn") { params.rope_scaling_type = LLAMA_ROPE_SCALING_TYPE_YARN; }
-        else { invalid_param = true; }
-        return true;
-    }
-    if (arg == "--rope-scale") {
-        CHECK_ARG
-        params.rope_freq_scale = 1.0f / std::stof(argv[i]);
-        return true;
-    }
-    if (arg == "--yarn-orig-ctx") {
-        CHECK_ARG
-        params.yarn_orig_ctx = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "--yarn-ext-factor") {
-        CHECK_ARG
-        params.yarn_ext_factor = std::stof(argv[i]);
-        return true;
-    }
-    if (arg == "--yarn-attn-factor") {
-        CHECK_ARG
-        params.yarn_attn_factor = std::stof(argv[i]);
-        return true;
-    }
-    if (arg == "--yarn-beta-fast") {
-        CHECK_ARG
-        params.yarn_beta_fast = std::stof(argv[i]);
-        return true;
-    }
-    if (arg == "--yarn-beta-slow") {
-        CHECK_ARG
-        params.yarn_beta_slow = std::stof(argv[i]);
-        return true;
-    }
-    if (arg == "--pooling") {
-        CHECK_ARG
-        std::string value(argv[i]);
-        /**/ if (value == "none") { params.pooling_type = LLAMA_POOLING_TYPE_NONE; }
-        else if (value == "mean") { params.pooling_type = LLAMA_POOLING_TYPE_MEAN; }
-        else if (value == "cls") { params.pooling_type = LLAMA_POOLING_TYPE_CLS; }
-        else if (value == "last") { params.pooling_type = LLAMA_POOLING_TYPE_LAST; }
-        else { invalid_param = true; }
-        return true;
-    }
-    if (arg == "--attention") {
-        CHECK_ARG
-        std::string value(argv[i]);
-        /**/ if (value == "causal") { params.attention_type = LLAMA_ATTENTION_TYPE_CAUSAL; }
-        else if (value == "non-causal") { params.attention_type = LLAMA_ATTENTION_TYPE_NON_CAUSAL; }
-        else { invalid_param = true; }
-        return true;
-    }
-    if (arg == "--defrag-thold" || arg == "-dt") {
-        CHECK_ARG
-        params.defrag_thold = std::stof(argv[i]);
-        return true;
-    }
-    if (arg == "--samplers") {
-        CHECK_ARG
-        const auto sampler_names = string_split(argv[i], ';');
-        sparams.samplers_sequence = llama_sampling_types_from_names(sampler_names, true);
-        return true;
-    }
-    if (arg == "--sampling-seq") {
-        CHECK_ARG
-        sparams.samplers_sequence = llama_sampling_types_from_chars(argv[i]);
-        return true;
-    }
-    if (arg == "--top-p") {
-        CHECK_ARG
-        sparams.top_p = std::stof(argv[i]);
-        return true;
-    }
-    if (arg == "--min-p") {
-        CHECK_ARG
-        sparams.min_p = std::stof(argv[i]);
-        return true;
-    }
-    if (arg == "--temp") {
-        CHECK_ARG
-        sparams.temp = std::stof(argv[i]);
-        sparams.temp = std::max(sparams.temp, 0.0f);
-        return true;
-    }
-    if (arg == "--tfs") {
-        CHECK_ARG
-        sparams.tfs_z = std::stof(argv[i]);
-        return true;
-    }
-    if (arg == "--typical") {
-        CHECK_ARG
-        sparams.typical_p = std::stof(argv[i]);
-        return true;
-    }
-    if (arg == "--repeat-last-n") {
-        CHECK_ARG
-        sparams.penalty_last_n = std::stoi(argv[i]);
-        sparams.n_prev = std::max(sparams.n_prev, sparams.penalty_last_n);
-        return true;
-    }
-    if (arg == "--repeat-penalty") {
-        CHECK_ARG
-        sparams.penalty_repeat = std::stof(argv[i]);
-        return true;
-    }
-    if (arg == "--frequency-penalty") {
-        CHECK_ARG
-        sparams.penalty_freq = std::stof(argv[i]);
-        return true;
-    }
-    if (arg == "--presence-penalty") {
-        CHECK_ARG
-        sparams.penalty_present = std::stof(argv[i]);
-        return true;
-    }
-    if (arg == "--dynatemp-range") {
-        CHECK_ARG
-        sparams.dynatemp_range = std::stof(argv[i]);
-        return true;
-    }
-    if (arg == "--dynatemp-exp") {
-        CHECK_ARG
-        sparams.dynatemp_exponent = std::stof(argv[i]);
-        return true;
-    }
-    if (arg == "--mirostat") {
-        CHECK_ARG
-        sparams.mirostat = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "--mirostat-lr") {
-        CHECK_ARG
-        sparams.mirostat_eta = std::stof(argv[i]);
-        return true;
-    }
-    if (arg == "--mirostat-ent") {
-        CHECK_ARG
-        sparams.mirostat_tau = std::stof(argv[i]);
-        return true;
-    }
-    if (arg == "--cfg-negative-prompt") {
-        CHECK_ARG
-        sparams.cfg_negative_prompt = argv[i];
-        return true;
-    }
-    if (arg == "--cfg-negative-prompt-file") {
-        CHECK_ARG
-        std::ifstream file(argv[i]);
-        if (!file) {
-            fprintf(stderr, "error: failed to open file '%s'\n", argv[i]);
-            invalid_param = true;
-            return true;
-        }
-        std::copy(std::istreambuf_iterator<char>(file), std::istreambuf_iterator<char>(), back_inserter(sparams.cfg_negative_prompt));
-        if (!sparams.cfg_negative_prompt.empty() && sparams.cfg_negative_prompt.back() == '\n') {
-            sparams.cfg_negative_prompt.pop_back();
-        }
-        return true;
-    }
-    if (arg == "--cfg-scale") {
-        CHECK_ARG
-        sparams.cfg_scale = std::stof(argv[i]);
-        return true;
-    }
-    if (arg == "-b" || arg == "--batch-size") {
-        CHECK_ARG
-        params.n_batch = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "-ub" || arg == "--ubatch-size") {
-        CHECK_ARG
-        params.n_ubatch = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "--keep") {
-        CHECK_ARG
-        params.n_keep = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "--draft") {
-        CHECK_ARG
-        params.n_draft = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "--chunks") {
-        CHECK_ARG
-        params.n_chunks = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "-np" || arg == "--parallel") {
-        CHECK_ARG
-        params.n_parallel = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "-ns" || arg == "--sequences") {
-        CHECK_ARG
-        params.n_sequences = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "--p-split" || arg == "-ps") {
-        CHECK_ARG
-        params.p_split = std::stof(argv[i]);
-        return true;
-    }
-    if (arg == "-m" || arg == "--model") {
-        CHECK_ARG
-        params.model = argv[i];
-        return true;
-    }
-    if (arg == "-md" || arg == "--model-draft") {
-        CHECK_ARG
-        params.model_draft = argv[i];
-        return true;
-    }
-    if (arg == "-a" || arg == "--alias") {
-        CHECK_ARG
-        params.model_alias = argv[i];
-        return true;
-    }
-    if (arg == "-mu" || arg == "--model-url") {
-        CHECK_ARG
-        params.model_url = argv[i];
-        return true;
-    }
-    if (arg == "-hft" || arg == "--hf-token") {
-        if (++i >= argc) {
-          invalid_param = true;
-          return true;
-        }
-        params.hf_token = argv[i];
-        return true;
-    }
-    if (arg == "-hfr" || arg == "--hf-repo") {
-        CHECK_ARG
-        params.hf_repo = argv[i];
-        return true;
-    }
-    if (arg == "-hff" || arg == "--hf-file") {
-        CHECK_ARG
-        params.hf_file = argv[i];
-        return true;
-    }
-    if (arg == "--lora") {
-        CHECK_ARG
-        params.lora_adapter.emplace_back(argv[i], 1.0f);
-        return true;
-    }
-    if (arg == "--lora-scaled") {
-        CHECK_ARG
-        const char* lora_adapter = argv[i];
-        CHECK_ARG
-        params.lora_adapter.emplace_back(lora_adapter, std::stof(argv[i]));
-        return true;
-    }
-    if (arg == "--control-vector") {
-        CHECK_ARG
-        params.control_vectors.push_back({ 1.0f, argv[i], });
-        return true;
-    }
-    if (arg == "--control-vector-scaled") {
-        CHECK_ARG
-        const char* fname = argv[i];
-        CHECK_ARG
-        params.control_vectors.push_back({ std::stof(argv[i]), fname, });
-        return true;
-    }
-    if (arg == "--control-vector-layer-range") {
-        CHECK_ARG
-        params.control_vector_layer_start = std::stoi(argv[i]);
-        CHECK_ARG
-        params.control_vector_layer_end = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "--mmproj") {
-        CHECK_ARG
-        params.mmproj = argv[i];
-        return true;
-    }
-    if (arg == "--image") {
-        CHECK_ARG
-        params.image.emplace_back(argv[i]);
-        return true;
-    }
-    if (arg == "-i" || arg == "--interactive") {
-        params.interactive = true;
-        return true;
-    }
-    if (arg == "-sp" || arg == "--special") {
-        params.special = true;
-        return true;
-    }
-    if (arg == "--embedding" || arg == "--embeddings") {
-        params.embedding = true;
-        return true;
-    }
-    if (arg == "--embd-normalize") {
-        CHECK_ARG
-        params.embd_normalize = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "--embd-output-format") {
-        CHECK_ARG
-        params.embd_out = argv[i];
-        return true;
-    }
-    if (arg == "--embd-separator") {
-        CHECK_ARG
-        params.embd_sep = argv[i];
-        return true;
-    }
-    if (arg == "-if" || arg == "--interactive-first") {
-        params.interactive_first = true;
-        return true;
-    }
-    if (arg == "-cnv" || arg == "--conversation") {
-        params.conversation = true;
-        return true;
-    }
-    if (arg == "--infill") {
-        params.infill = true;
-        return true;
-    }
-    if (arg == "-dkvc" || arg == "--dump-kv-cache") {
-        params.dump_kv_cache = true;
-        return true;
-    }
-    if (arg == "-nkvo" || arg == "--no-kv-offload") {
-        params.no_kv_offload = true;
-        return true;
-    }
-    if (arg == "-ctk" || arg == "--cache-type-k") {
-        params.cache_type_k = argv[++i];
-        return true;
-    }
-    if (arg == "-ctv" || arg == "--cache-type-v") {
-        params.cache_type_v = argv[++i];
-        return true;
-    }
-    if (arg == "-mli" || arg == "--multiline-input") {
-        params.multiline_input = true;
-        return true;
-    }
-    if (arg == "--simple-io") {
-        params.simple_io = true;
-        return true;
-    }
-    if (arg == "-cb" || arg == "--cont-batching") {
-        params.cont_batching = true;
-        return true;
-    }
-    if (arg == "-nocb" || arg == "--no-cont-batching") {
-        params.cont_batching = false;
-        return true;
-    }
-    if (arg == "-fa" || arg == "--flash-attn") {
-        params.flash_attn = true;
-        return true;
-    }
-    if (arg == "-co" || arg == "--color") {
-        params.use_color = true;
-        return true;
-    }
-    if (arg == "--mlock") {
-        params.use_mlock = true;
-        return true;
-    }
-    if (arg == "-ngl" || arg == "--gpu-layers" || arg == "--n-gpu-layers") {
-        CHECK_ARG
-        params.n_gpu_layers = std::stoi(argv[i]);
-        if (!llama_supports_gpu_offload()) {
-            fprintf(stderr, "warning: not compiled with GPU offload support, --gpu-layers option will be ignored\n");
-            fprintf(stderr, "warning: see main README.md for information on enabling GPU BLAS support\n");
-        }
-        return true;
-    }
-    if (arg == "-ngld" || arg == "--gpu-layers-draft" || arg == "--gpu-layers-draft") {
-        CHECK_ARG
-        params.n_gpu_layers_draft = std::stoi(argv[i]);
-        if (!llama_supports_gpu_offload()) {
-            fprintf(stderr, "warning: not compiled with GPU offload support, --gpu-layers-draft option will be ignored\n");
-            fprintf(stderr, "warning: see main README.md for information on enabling GPU BLAS support\n");
-        }
-        return true;
-    }
-    if (arg == "--main-gpu" || arg == "-mg") {
-        CHECK_ARG
-        params.main_gpu = std::stoi(argv[i]);
-#ifndef GGML_USE_CUDA_SYCL_VULKAN
-        fprintf(stderr, "warning: llama.cpp was compiled without CUDA/SYCL/Vulkan. Setting the main GPU has no effect.\n");
-#endif // GGML_USE_CUDA_SYCL_VULKAN
-        return true;
-    }
-    if (arg == "--split-mode" || arg == "-sm") {
-        CHECK_ARG
-        std::string arg_next = argv[i];
-        if (arg_next == "none") {
-            params.split_mode = LLAMA_SPLIT_MODE_NONE;
-        }
-        else if (arg_next == "layer") {
-            params.split_mode = LLAMA_SPLIT_MODE_LAYER;
-        }
-        else if (arg_next == "row") {
-#ifdef GGML_USE_SYCL
-            fprintf(stderr, "warning: The split mode value:[row] is not supported by llama.cpp with SYCL. It's developing.\nExit!\n");
-            exit(1);
-#endif // GGML_USE_SYCL
-            params.split_mode = LLAMA_SPLIT_MODE_ROW;
-        }
-        else {
-            invalid_param = true;
-            return true;
-        }
-#ifndef GGML_USE_CUDA_SYCL_VULKAN
-        fprintf(stderr, "warning: llama.cpp was compiled without CUDA/SYCL/Vulkan. Setting the split mode has no effect.\n");
-#endif // GGML_USE_CUDA_SYCL_VULKAN
-        return true;
-    }
-    if (arg == "--tensor-split" || arg == "-ts") {
-        CHECK_ARG
-        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<std::string> split_arg{ it, {} };
-        if (split_arg.size() >= llama_max_devices()) {
-            invalid_param = true;
-            return true;
-        }
-        for (size_t i = 0; i < llama_max_devices(); ++i) {
-            if (i < split_arg.size()) {
-                params.tensor_split[i] = std::stof(split_arg[i]);
-            }
-            else {
-                params.tensor_split[i] = 0.0f;
-            }
-        }
-#ifndef GGML_USE_CUDA_SYCL_VULKAN
-        fprintf(stderr, "warning: llama.cpp was compiled without CUDA/SYCL/Vulkan. Setting a tensor split has no effect.\n");
-#endif // GGML_USE_CUDA_SYCL_VULKAN
-        return true;
-    }
-    if (arg == "--rpc") {
-        CHECK_ARG
-        params.rpc_servers = argv[i];
-        return true;
-    }
-    if (arg == "--no-mmap") {
-        params.use_mmap = false;
-        return true;
-    }
-    if (arg == "--numa") {
-        CHECK_ARG
-        std::string value(argv[i]);
-        /**/ if (value == "distribute" || value == "") { params.numa = GGML_NUMA_STRATEGY_DISTRIBUTE; }
-        else if (value == "isolate") { params.numa = GGML_NUMA_STRATEGY_ISOLATE; }
-        else if (value == "numactl") { params.numa = GGML_NUMA_STRATEGY_NUMACTL; }
-        else { invalid_param = true; }
-        return true;
-    }
-    if (arg == "-v" || arg == "--verbose") {
-        params.verbosity = 1;
-        return true;
-    }
-    if (arg == "--verbosity") {
-        CHECK_ARG
-        params.verbosity = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "--verbose-prompt") {
-        params.verbose_prompt = true;
-        return true;
-    }
-    if (arg == "--no-display-prompt") {
-        params.display_prompt = false;
-        return true;
-    }
-    if (arg == "-r" || arg == "--reverse-prompt") {
-        CHECK_ARG
-        params.antiprompt.emplace_back(argv[i]);
-        return true;
-    }
-    if (arg == "-ld" || arg == "--logdir") {
-        CHECK_ARG
-        params.logdir = argv[i];
-
-        if (params.logdir.back() != DIRECTORY_SEPARATOR) {
-            params.logdir += DIRECTORY_SEPARATOR;
-        }
-        return true;
-    }
-    if (arg == "-lcs" || arg == "--lookup-cache-static") {
-        CHECK_ARG
-        params.lookup_cache_static = argv[i];
-        return true;
-    }
-    if (arg == "-lcd" || arg == "--lookup-cache-dynamic") {
-        CHECK_ARG
-        params.lookup_cache_dynamic = argv[i];
-        return true;
-    }
-    if (arg == "--save-all-logits" || arg == "--kl-divergence-base") {
-        CHECK_ARG
-        params.logits_file = argv[i];
-        return true;
-    }
-    if (arg == "--perplexity" || arg == "--all-logits") {
-        params.logits_all = true;
-        return true;
-    }
-    if (arg == "--ppl-stride") {
-        CHECK_ARG
-        params.ppl_stride = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "--ppl-output-type") {
-        CHECK_ARG
-        params.ppl_output_type = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "-ptc" || arg == "--print-token-count") {
-        CHECK_ARG
-        params.n_print = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "--check-tensors") {
-        params.check_tensors = true;
-        return true;
-    }
-    if (arg == "--hellaswag") {
-        params.hellaswag = true;
-        return true;
-    }
-    if (arg == "--hellaswag-tasks") {
-        CHECK_ARG
-        params.hellaswag_tasks = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "--winogrande") {
-        params.winogrande = true;
-        return true;
-    }
-    if (arg == "--winogrande-tasks") {
-        CHECK_ARG
-        params.winogrande_tasks = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "--multiple-choice") {
-        params.multiple_choice = true;
-        return true;
-    }
-    if (arg == "--multiple-choice-tasks") {
-        CHECK_ARG
-        params.multiple_choice_tasks = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "--kl-divergence") {
-        params.kl_divergence = true;
-        return true;
-    }
-    if (arg == "--ignore-eos") {
-        params.ignore_eos = true;
-        return true;
-    }
-    if (arg == "--penalize-nl") {
-        sparams.penalize_nl = true;
-        return true;
-    }
-    if (arg == "-l" || arg == "--logit-bias") {
-        CHECK_ARG
-        std::stringstream ss(argv[i]);
-        llama_token key;
-        char sign;
-        std::string value_str;
-        try {
-            if (ss >> key && ss >> sign && std::getline(ss, value_str) && (sign == '+' || sign == '-')) {
-                sparams.logit_bias[key] = std::stof(value_str) * ((sign == '-') ? -1.0f : 1.0f);
-            }
-            else {
-                throw std::exception();
-            }
-        }
-        catch (const std::exception&) {
-            invalid_param = true;
-            return true;
-        }
-        return true;
-    }
-    if (arg == "-h" || arg == "--help" || arg == "--usage"  ) {
-        params.usage = true;
-        return true;
-    }
-    if (arg == "--version") {
-        fprintf(stderr, "version: %d (%s)\n", LLAMA_BUILD_NUMBER, LLAMA_COMMIT);
-        fprintf(stderr, "built with %s for %s\n", LLAMA_COMPILER, LLAMA_BUILD_TARGET);
-        exit(0);
-    }
-    if (arg == "--in-prefix-bos") {
-        params.input_prefix_bos = true;
-        params.enable_chat_template = false;
-        return true;
-    }
-    if (arg == "--in-prefix") {
-        CHECK_ARG
-        params.input_prefix = argv[i];
-        params.enable_chat_template = false;
-        return true;
-    }
-    if (arg == "--in-suffix") {
-        CHECK_ARG
-        params.input_suffix = argv[i];
-        params.enable_chat_template = false;
-        return true;
-    }
-    if (arg == "--spm-infill") {
-        params.spm_infill = true;
-        return true;
-    }
-    if (arg == "--grammar") {
-        CHECK_ARG
-        sparams.grammar = argv[i];
-        return true;
-    }
-    if (arg == "--grammar-file") {
-        CHECK_ARG
-        std::ifstream file(argv[i]);
-        if (!file) {
-            fprintf(stderr, "error: failed to open file '%s'\n", argv[i]);
-            invalid_param = true;
-            return true;
-        }
-        std::copy(
-            std::istreambuf_iterator<char>(file),
-            std::istreambuf_iterator<char>(),
-            std::back_inserter(sparams.grammar)
-        );
-        return true;
-    }
-    if (arg == "-j" || arg == "--json-schema") {
-        CHECK_ARG
-        sparams.grammar = json_schema_to_grammar(json::parse(argv[i]));
-        return true;
-    }
-    if (arg == "--override-kv") {
-        CHECK_ARG
-        if (!string_parse_kv_override(argv[i], params.kv_overrides)) {
-            fprintf(stderr, "error: Invalid type for KV override: %s\n", argv[i]);
-            invalid_param = true;
-            return true;
-        }
-        return true;
-    }
-    if (arg == "--host") {
-        CHECK_ARG
-        params.hostname = argv[i];
-        return true;
-    }
-    if (arg == "--port") {
-        CHECK_ARG
-        params.port = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "--path") {
-        CHECK_ARG
-        params.public_path = argv[i];
-        return true;
+        if (is_running_on_efficiency_core()) {
+            continue; // efficiency cores harm lockstep threading
+        }
+        ++cpu; // hyperthreading isn't useful for linear algebra
+        ++result;
     }
-    if (arg == "--api-key") {
-        CHECK_ARG
-        params.api_keys.push_back(argv[i]);
-        return true;
+    return result;
+}
+
+#endif // __x86_64__ && __linux__
+
+/**
+ * Returns number of CPUs on system that are useful for math.
+ */
+int32_t cpu_get_num_math() {
+#if defined(__x86_64__) && defined(__linux__) && !defined(__ANDROID__)
+    int n_cpu = sysconf(_SC_NPROCESSORS_ONLN);
+    if (n_cpu < 1) {
+        return cpu_get_num_physical_cores();
     }
-    if (arg == "--api-key-file") {
-        CHECK_ARG
-        std::ifstream key_file(argv[i]);
-        if (!key_file) {
-            fprintf(stderr, "error: failed to open file '%s'\n", argv[i]);
-            invalid_param = true;
-            return true;
-        }
-        std::string key;
-        while (std::getline(key_file, key)) {
-            if (!key.empty()) {
-                params.api_keys.push_back(key);
+    if (is_hybrid_cpu()) {
+        cpu_set_t affinity;
+        if (!pthread_getaffinity_np(pthread_self(), sizeof(affinity), &affinity)) {
+            int result = cpu_count_math_cpus(n_cpu);
+            pthread_setaffinity_np(pthread_self(), sizeof(affinity), &affinity);
+            if (result > 0) {
+                return result;
             }
         }
-        key_file.close();
-        return true;
-    }
-    if (arg == "--ssl-key-file") {
-        CHECK_ARG
-        params.ssl_file_key = argv[i];
-        return true;
-    }
-    if (arg == "--ssl-cert-file") {
-        CHECK_ARG
-        params.ssl_file_cert = argv[i];
-        return true;
     }
-    if (arg == "--timeout" || arg == "-to") {
-        CHECK_ARG
-        params.timeout_read  = std::stoi(argv[i]);
-        params.timeout_write = std::stoi(argv[i]);
+#endif
+    return cpu_get_num_physical_cores();
+}
+
+// Helper for setting process priority
+
+#if defined(_WIN32)
+
+bool set_process_priority(enum ggml_sched_priority prio) {
+    if (prio == GGML_SCHED_PRIO_NORMAL) {
         return true;
     }
-    if (arg == "--threads-http") {
-        CHECK_ARG
-        params.n_threads_http = std::stoi(argv[i]);
-        return true;
+
+    DWORD p = NORMAL_PRIORITY_CLASS;
+    switch (prio) {
+        case GGML_SCHED_PRIO_NORMAL:   p = NORMAL_PRIORITY_CLASS;       break;
+        case GGML_SCHED_PRIO_MEDIUM:   p = ABOVE_NORMAL_PRIORITY_CLASS; break;
+        case GGML_SCHED_PRIO_HIGH:     p = HIGH_PRIORITY_CLASS;         break;
+        case GGML_SCHED_PRIO_REALTIME: p = REALTIME_PRIORITY_CLASS;     break;
     }
-    if (arg == "-spf" || arg == "--system-prompt-file") {
-        CHECK_ARG
-        std::ifstream file(argv[i]);
-        if (!file) {
-            fprintf(stderr, "error: failed to open file '%s'\n", argv[i]);
-            invalid_param = true;
-            return true;
-        }
-        std::string system_prompt;
-        std::copy(
-                std::istreambuf_iterator<char>(file),
-                std::istreambuf_iterator<char>(),
-                std::back_inserter(system_prompt)
-                );
-        params.system_prompt = system_prompt;
-        return true;
+
+    if (!SetPriorityClass(GetCurrentProcess(), p)) {
+        LOG_WRN("failed to set process priority class %d : (%d)\n", prio, (int) GetLastError());
+        return false;
     }
-    if (arg == "--log-format") {
-        CHECK_ARG
-        if (std::strcmp(argv[i], "json") == 0) {
-            params.log_json = true;
-        } else if (std::strcmp(argv[i], "text") == 0) {
-            params.log_json = false;
-        } else {
-            invalid_param = true;
-            return true;
-        }
+
+    return true;
+}
+
+#else // MacOS and POSIX
+#include <sys/types.h>
+#include <sys/resource.h>
+
+bool set_process_priority(enum ggml_sched_priority prio) {
+    if (prio == GGML_SCHED_PRIO_NORMAL) {
         return true;
     }
-    if (arg == "--no-slots") {
-        params.endpoint_slots = false;
-        return true;
+
+    int p = 0;
+    switch (prio) {
+        case GGML_SCHED_PRIO_NORMAL:   p =  0;  break;
+        case GGML_SCHED_PRIO_MEDIUM:   p = -5;  break;
+        case GGML_SCHED_PRIO_HIGH:     p = -10; break;
+        case GGML_SCHED_PRIO_REALTIME: p = -20; break;
     }
-    if (arg == "--metrics") {
-        params.endpoint_metrics = true;
-        return true;
+
+    if (!setpriority(PRIO_PROCESS, 0, p)) {
+        LOG_WRN("failed to set process priority %d : %s (%d)\n", prio, strerror(errno), errno);
+        return false;
     }
-    if (arg == "--slot-save-path") {
-        CHECK_ARG
-        params.slot_save_path = argv[i];
-        // if doesn't end with DIRECTORY_SEPARATOR, add it
-        if (!params.slot_save_path.empty() && params.slot_save_path[params.slot_save_path.size() - 1] != DIRECTORY_SEPARATOR) {
-            params.slot_save_path += DIRECTORY_SEPARATOR;
+    return true;
+}
+
+#endif
+
+//
+// CLI argument parsing
+//
+
+
+void postprocess_cpu_params(cpu_params& cpuparams, const cpu_params* role_model) {
+    int32_t n_set = 0;
+
+    if (cpuparams.n_threads < 0) {
+        // Assuming everything about cpuparams is invalid
+        if (role_model != nullptr) {
+            cpuparams = *role_model;
+        } else {
+            cpuparams.n_threads = cpu_get_num_math();
         }
-        return true;
     }
-    if (arg == "--chat-template") {
-        CHECK_ARG
-        if (!llama_chat_verify_template(argv[i])) {
-            fprintf(stderr, "error: the supplied chat template is not supported: %s\n", argv[i]);
-            fprintf(stderr, "note: llama.cpp does not use jinja parser, we only support commonly used templates\n");
-            invalid_param = true;
-            return true;
+
+    for (int32_t i = 0; i < GGML_MAX_N_THREADS; i++) {
+        if (cpuparams.cpumask[i]) {
+            n_set++;
         }
-        params.chat_template = argv[i];
-        return true;
-    }
-    if (arg == "--slot-prompt-similarity" || arg == "-sps") {
-        CHECK_ARG
-        params.slot_prompt_similarity = std::stof(argv[i]);
-        return true;
-    }
-    if (arg == "-pps") {
-        params.is_pp_shared = true;
-        return true;
     }
-    if (arg == "-npp") {
-        CHECK_ARG
-        auto p = string_split<int>(argv[i], split_delim);
-        params.n_pp.insert(params.n_pp.end(), p.begin(), p.end());
-        return true;
-    }
-    if (arg == "-ntg") {
-        CHECK_ARG
-        auto p = string_split<int>(argv[i], split_delim);
-        params.n_tg.insert(params.n_tg.end(), p.begin(), p.end());
-        return true;
+
+    if (n_set && n_set < cpuparams.n_threads) {
+        // Not enough set bits, may experience performance issues.
+        LOG_WRN("Not enough set bits in CPU mask (%d) to satisfy requested thread count: %d\n", n_set, cpuparams.n_threads);
     }
-    if (arg == "-npl") {
-        CHECK_ARG
-        auto p = string_split<int>(argv[i], split_delim);
-        params.n_pl.insert(params.n_pl.end(), p.begin(), p.end());
-        return true;
+}
+
+bool parse_cpu_range(const std::string & range, bool (&boolmask)[GGML_MAX_N_THREADS]) {
+    size_t dash_loc = range.find('-');
+    if (dash_loc == std::string::npos) {
+        LOG_ERR("Format of CPU range is invalid! Expected [<start>]-[<end>].\n");
+        return false;
     }
-    if (arg == "--context-file") {
-        CHECK_ARG
-        std::ifstream file(argv[i], std::ios::binary);
-        if (!file) {
-            fprintf(stderr, "error: failed to open file '%s'\n", argv[i]);
-            invalid_param = true;
-            return true;
+
+    size_t start_i;
+    size_t end_i;
+
+    if (dash_loc == 0) {
+        start_i = 0;
+    } else {
+        start_i = std::stoull(range.substr(0, dash_loc));
+        if (start_i >= GGML_MAX_N_THREADS) {
+            LOG_ERR("Start index out of bounds!\n");
+            return false;
         }
-        params.context_files.push_back(argv[i]);
-        return true;
-    }
-    if (arg == "--chunk-size") {
-        CHECK_ARG
-        params.chunk_size = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "--chunk-separator") {
-        CHECK_ARG
-        params.chunk_separator = argv[i];
-        return true;
-    }
-    if (arg == "--junk") {
-        CHECK_ARG
-        params.n_junk = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "--pos") {
-        CHECK_ARG
-        params.i_pos = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "-o" || arg == "--output" || arg == "--output-file") {
-        CHECK_ARG
-        params.out_file = argv[i];
-        params.cvector_outfile = argv[i];
-        params.lora_outfile = argv[i];
-        return true;
-    }
-    if (arg == "-ofreq" || arg == "--output-frequency") {
-        CHECK_ARG
-        params.n_out_freq = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "--save-frequency") {
-        CHECK_ARG
-        params.n_save_freq = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "--process-output") {
-        params.process_output = true;
-        return true;
-    }
-    if (arg == "--no-ppl") {
-        params.compute_ppl = false;
-        return true;
-    }
-    if (arg == "--chunk" || arg == "--from-chunk") {
-        CHECK_ARG
-        params.i_chunk = std::stoi(argv[i]);
-        return true;
-    }
-    // cvector params
-    if (arg == "--positive-file") {
-        CHECK_ARG
-        params.cvector_positive_file = argv[i];
-        return true;
-    }
-    if (arg == "--negative-file") {
-        CHECK_ARG
-        params.cvector_negative_file = argv[i];
-        return true;
-    }
-    if (arg == "--pca-batch") {
-        CHECK_ARG
-        params.n_pca_batch = std::stoi(argv[i]);
-        return true;
     }
-    if (arg == "--pca-iter") {
-        CHECK_ARG
-        params.n_pca_iterations = std::stoi(argv[i]);
-        return true;
-    }
-    if (arg == "--method") {
-        CHECK_ARG
-        std::string value(argv[i]);
-        /**/ if (value == "pca") { params.cvector_dimre_method = DIMRE_METHOD_PCA; }
-        else if (value == "mean") { params.cvector_dimre_method = DIMRE_METHOD_MEAN; }
-        else { invalid_param = true; }
-        return true;
-    }
-    if (arg == "--no-warmup") {
-        params.warmup = false;
-        return true;
-    }
-#ifndef LOG_DISABLE_LOGS
-    // Parse args for logging parameters
-    if (log_param_single_parse(argv[i])) {
-        // Do nothing, log_param_single_parse automatically does it's thing
-        //  and returns if a match was found and parsed.
-        return true;
-    }
-    if (log_param_pair_parse( /*check_but_dont_parse*/ true, argv[i])) {
-        // We have a matching known parameter requiring an argument,
-        //  now we need to check if there is anything after this argv
-        //  and flag invalid_param or parse it.
-        CHECK_ARG
-        if (!log_param_pair_parse( /*check_but_dont_parse*/ false, argv[i - 1], argv[i])) {
-            invalid_param = true;
-            return true;
+
+    if (dash_loc == range.length() - 1) {
+        end_i = GGML_MAX_N_THREADS - 1;
+    } else {
+        end_i = std::stoull(range.substr(dash_loc + 1));
+        if (end_i >= GGML_MAX_N_THREADS) {
+            LOG_ERR("End index out of bounds!\n");
+            return false;
         }
-        return true;
     }
-    // End of Parse args for logging parameters
-#endif // LOG_DISABLE_LOGS
 
-    return false;
+    for (size_t i = start_i; i <= end_i; i++) {
+        boolmask[i] = true;
+    }
+
+    return true;
 }
 
-#ifdef __GNUC__
-#ifdef __MINGW32__
-#define LLAMA_COMMON_ATTRIBUTE_FORMAT(...) __attribute__((format(gnu_printf, __VA_ARGS__)))
-#else
-#define LLAMA_COMMON_ATTRIBUTE_FORMAT(...) __attribute__((format(printf, __VA_ARGS__)))
-#endif
-#else
-#define LLAMA_COMMON_ATTRIBUTE_FORMAT(...)
-#endif
+bool parse_cpu_mask(const std::string & mask, bool (&boolmask)[GGML_MAX_N_THREADS]) {
+    // Discard potential 0x prefix
+    size_t start_i = 0;
+    if (mask.length() >= 2 && mask.substr(0, 2) == "0x") {
+        start_i = 2;
+    }
 
-void gpt_params_print_usage(int /*argc*/, char ** argv, const gpt_params & params) {
-    const llama_sampling_params & sparams = params.sparams;
+    size_t num_digits = mask.length() - start_i;
+    if (num_digits > 128) num_digits = 128;
 
-    std::string sampler_type_chars;
-    std::string sampler_type_names;
-    for (const auto sampler_type : sparams.samplers_sequence) {
-        sampler_type_chars += static_cast<char>(sampler_type);
-        sampler_type_names += llama_sampling_type_to_str(sampler_type) + ";";
-    }
-    sampler_type_names.pop_back();
+    size_t end_i = num_digits + start_i;
+
+    for (size_t i = start_i, n = (num_digits*4 - 1); i < end_i; i++, n-=4) {
+        char c = mask.at(i);
+        int8_t id = c;
 
-    struct option_info {
-        LLAMA_COMMON_ATTRIBUTE_FORMAT(4, 5)
-        option_info(const std::string & tags, const char * args, const char * desc, ...) : tags(tags), args(args), desc(desc) {
-            va_list args_list;
-            va_start(args_list, desc);
-            char buffer[1024];
-            vsnprintf(buffer, sizeof(buffer), desc, args_list);
-            va_end(args_list);
-            this->desc = buffer;
+        if ((c >= '0' && c <= '9')) {
+            id -= '0';
+        } else if (c >= 'a' && c <= 'f') {
+            id -= 'a' - 10;
+        } else if (c >= 'A' && c <= 'F') {
+            id -= 'A' - 10;
+        } else {
+            LOG_ERR("Invalid hex character '%c' at position %d\n", c, int32_t(i));
+            return false;
         }
 
-        option_info(const std::string & grp) : grp(grp) {}
+        boolmask[  n  ] = boolmask[  n  ] || ((id & 8) != 0);
+        boolmask[n - 1] = boolmask[n - 1] || ((id & 4) != 0);
+        boolmask[n - 2] = boolmask[n - 2] || ((id & 2) != 0);
+        boolmask[n - 3] = boolmask[n - 3] || ((id & 1) != 0);
+    }
 
-        std::string tags;
-        std::string args;
-        std::string desc;
-        std::string grp;
-    };
+    return true;
+}
+
+void gpt_init() {
+    llama_log_set([](ggml_log_level level, const char * text, void * /*user_data*/) {
+        if (LOG_DEFAULT_LLAMA <= gpt_log_verbosity_thold) {
+            gpt_log_add(gpt_log_main(), level, "%s", text);
+        }
+    }, NULL);
 
-    std::vector<option_info> options;
-
-    // TODO: filter by tags
-
-    options.push_back({ "general" });
-    options.push_back({ "*",           "-h,    --help, --usage",        "print usage and exit" });
-    options.push_back({ "*",           "       --version",              "show version and build info" });
-    options.push_back({ "*",           "-v,    --verbose",              "print verbose information" });
-    options.push_back({ "*",           "       --verbosity N",          "set specific verbosity level (default: %d)", params.verbosity });
-    options.push_back({ "*",           "       --verbose-prompt",       "print a verbose prompt before generation (default: %s)", params.verbose_prompt ? "true" : "false" });
-    options.push_back({ "*",           "       --no-display-prompt",    "don't print prompt at generation (default: %s)", !params.display_prompt ? "true" : "false" });
-    options.push_back({ "*",           "-co,   --color",                "colorise output to distinguish prompt and user input from generations (default: %s)", params.use_color ? "true" : "false" });
-    options.push_back({ "*",           "-s,    --seed SEED",            "RNG seed (default: %d, use random seed for < 0)", params.seed });
-    options.push_back({ "*",           "-t,    --threads N",            "number of threads to use during generation (default: %d)", params.n_threads });
-    options.push_back({ "*",           "-tb,   --threads-batch N",      "number of threads to use during batch and prompt processing (default: same as --threads)" });
-    options.push_back({ "speculative", "-td,   --threads-draft N",      "number of threads to use during generation (default: same as --threads)" });
-    options.push_back({ "speculative", "-tbd,  --threads-batch-draft N",
-                                                                        "number of threads to use during batch and prompt processing (default: same as --threads-draft)" });
-    options.push_back({ "speculative", "       --draft N",              "number of tokens to draft for speculative decoding (default: %d)", params.n_draft });
-    options.push_back({ "speculative", "-ps,   --p-split N",            "speculative decoding split probability (default: %.1f)", (double)params.p_split });
-    options.push_back({ "*",           "-lcs,  --lookup-cache-static FNAME",
-                                                                        "path to static lookup cache to use for lookup decoding (not updated by generation)" });
-    options.push_back({ "*",           "-lcd,  --lookup-cache-dynamic FNAME",
-                                                                        "path to dynamic lookup cache to use for lookup decoding (updated by generation)" });
-
-    options.push_back({ "*",           "-c,    --ctx-size N",           "size of the prompt context (default: %d, 0 = loaded from model)", params.n_ctx });
-    options.push_back({ "*",           "-n,    --predict N",            "number of tokens to predict (default: %d, -1 = infinity, -2 = until context filled)", params.n_predict });
-    options.push_back({ "*",           "-b,    --batch-size N",         "logical maximum batch size (default: %d)", params.n_batch });
-    options.push_back({ "*",           "-ub,   --ubatch-size N",        "physical maximum batch size (default: %d)", params.n_ubatch });
-    options.push_back({ "*",           "       --keep N",               "number of tokens to keep from the initial prompt (default: %d, -1 = all)", params.n_keep });
-    options.push_back({ "*",           "       --chunks N",             "max number of chunks to process (default: %d, -1 = all)", params.n_chunks });
-    options.push_back({ "*",           "-fa,   --flash-attn",           "enable Flash Attention (default: %s)", params.flash_attn ? "enabled" : "disabled" });
-    options.push_back({ "*",           "-p,    --prompt PROMPT",        "prompt to start generation with\n"
-                                                                        "in conversation mode, this will be used as system prompt\n"
-                                                                        "(default: '%s')", params.prompt.c_str() });
-    options.push_back({ "*",           "-f,    --file FNAME",           "a file containing the prompt (default: none)" });
-    options.push_back({ "*",           "       --in-file FNAME",        "an input file (repeat to specify multiple files)" });
-    options.push_back({ "*",           "-bf,   --binary-file FNAME",    "binary file containing the prompt (default: none)" });
-    options.push_back({ "*",           "-e,    --escape",               "process escapes sequences (\\n, \\r, \\t, \\', \\\", \\\\) (default: %s)", params.escape ? "true" : "false" });
-    options.push_back({ "*",           "       --no-escape",            "do not process escape sequences" });
-    options.push_back({ "main",        "-ptc,  --print-token-count N",  "print token count every N tokens (default: %d)", params.n_print });
-    options.push_back({ "main",        "       --prompt-cache FNAME",   "file to cache prompt state for faster startup (default: none)" });
-    options.push_back({ "main",        "       --prompt-cache-all",     "if specified, saves user input and generations to cache as well\n"
-                                                                        "not supported with --interactive or other interactive options" });
-    options.push_back({ "main",        "       --prompt-cache-ro",      "if specified, uses the prompt cache but does not update it" });
-    options.push_back({ "main",        "-r,    --reverse-prompt PROMPT",
-                                                                        "halt generation at PROMPT, return control in interactive mode\n"
-                                                                        "can be specified more than once for multiple prompts" });
-    options.push_back({ "main",        "-sp,   --special",              "special tokens output enabled (default: %s)", params.special ? "true" : "false" });
-    options.push_back({ "main",        "-cnv,  --conversation",         "run in conversation mode, does not print special tokens and suffix/prefix\n"
-                                                                        "if suffix/prefix are not specified, default chat template will be used\n"
-                                                                        "(default: %s)", params.conversation ? "true" : "false" });
-    options.push_back({ "main infill", "-i,    --interactive",          "run in interactive mode (default: %s)", params.interactive ? "true" : "false" });
-    options.push_back({ "main infill", "-if,   --interactive-first",    "run in interactive mode and wait for input right away (default: %s)", params.interactive_first ? "true" : "false" });
-    options.push_back({ "main infill", "-mli,  --multiline-input",      "allows you to write or paste multiple lines without ending each in '\\'" });
-    options.push_back({ "main infill", "       --in-prefix-bos",        "prefix BOS to user inputs, preceding the `--in-prefix` string" });
-    options.push_back({ "main infill", "       --in-prefix STRING",     "string to prefix user inputs with (default: empty)" });
-    options.push_back({ "main infill", "       --in-suffix STRING",     "string to suffix after user inputs with (default: empty)" });
-    options.push_back({ "main",        "       --no-warmup",            "skip warming up the model with an empty run" });
-    options.push_back({ "server infill",
-                                       "       --spm-infill",           "use Suffix/Prefix/Middle pattern for infill (instead of Prefix/Suffix/Middle) as some models prefer this. (default: %s)", params.spm_infill ? "enabled" : "disabled" });
-
-    options.push_back({ "sampling" });
-    options.push_back({ "*",           "       --samplers SAMPLERS",    "samplers that will be used for generation in the order, separated by \';\'\n"
-                                                                        "(default: %s)", sampler_type_names.c_str() });
-    options.push_back({ "*",           "       --sampling-seq SEQUENCE",
-                                                                        "simplified sequence for samplers that will be used (default: %s)", sampler_type_chars.c_str() });
-    options.push_back({ "*",           "       --ignore-eos",           "ignore end of stream token and continue generating (implies --logit-bias EOS-inf)" });
-    options.push_back({ "*",           "       --penalize-nl",          "penalize newline tokens (default: %s)", sparams.penalize_nl ? "true" : "false" });
-    options.push_back({ "*",           "       --temp N",               "temperature (default: %.1f)", (double)sparams.temp });
-    options.push_back({ "*",           "       --top-k N",              "top-k sampling (default: %d, 0 = disabled)", sparams.top_k });
-    options.push_back({ "*",           "       --top-p N",              "top-p sampling (default: %.1f, 1.0 = disabled)", (double)sparams.top_p });
-    options.push_back({ "*",           "       --min-p N",              "min-p sampling (default: %.1f, 0.0 = disabled)", (double)sparams.min_p });
-    options.push_back({ "*",           "       --tfs N",                "tail free sampling, parameter z (default: %.1f, 1.0 = disabled)", (double)sparams.tfs_z });
-    options.push_back({ "*",           "       --typical N",            "locally typical sampling, parameter p (default: %.1f, 1.0 = disabled)", (double)sparams.typical_p });
-    options.push_back({ "*",           "       --repeat-last-n N",      "last n tokens to consider for penalize (default: %d, 0 = disabled, -1 = ctx_size)", sparams.penalty_last_n });
-    options.push_back({ "*",           "       --repeat-penalty N",     "penalize repeat sequence of tokens (default: %.1f, 1.0 = disabled)", (double)sparams.penalty_repeat });
-    options.push_back({ "*",           "       --presence-penalty N",   "repeat alpha presence penalty (default: %.1f, 0.0 = disabled)", (double)sparams.penalty_present });
-    options.push_back({ "*",           "       --frequency-penalty N",  "repeat alpha frequency penalty (default: %.1f, 0.0 = disabled)", (double)sparams.penalty_freq });
-    options.push_back({ "*",           "       --dynatemp-range N",     "dynamic temperature range (default: %.1f, 0.0 = disabled)", (double)sparams.dynatemp_range });
-    options.push_back({ "*",           "       --dynatemp-exp N",       "dynamic temperature exponent (default: %.1f)", (double)sparams.dynatemp_exponent });
-    options.push_back({ "*",           "       --mirostat N",           "use Mirostat sampling.\n"
-                                                                        "Top K, Nucleus, Tail Free and Locally Typical samplers are ignored if used.\n"
-                                                                        "(default: %d, 0 = disabled, 1 = Mirostat, 2 = Mirostat 2.0)", sparams.mirostat });
-    options.push_back({ "*",           "       --mirostat-lr N",        "Mirostat learning rate, parameter eta (default: %.1f)", (double)sparams.mirostat_eta });
-    options.push_back({ "*",           "       --mirostat-ent N",       "Mirostat target entropy, parameter tau (default: %.1f)", (double)sparams.mirostat_tau });
-    options.push_back({ "*",           "       -l TOKEN_ID(+/-)BIAS",   "modifies the likelihood of token appearing in the completion,\n"
-                                                                        "i.e. `--logit-bias 15043+1` to increase likelihood of token ' Hello',\n"
-                                                                        "or `--logit-bias 15043-1` to decrease likelihood of token ' Hello'" });
-    options.push_back({ "main",        "       --cfg-negative-prompt PROMPT",
-                                                                        "negative prompt to use for guidance (default: '%s')", sparams.cfg_negative_prompt.c_str() });
-    options.push_back({ "main",        "       --cfg-negative-prompt-file FNAME",
-                                                                        "negative prompt file to use for guidance" });
-    options.push_back({ "main",        "       --cfg-scale N",          "strength of guidance (default: %.1f, 1.0 = disable)", (double)sparams.cfg_scale });
-    options.push_back({ "main",        "       --chat-template JINJA_TEMPLATE",
-                                                                        "set custom jinja chat template (default: template taken from model's metadata)\n"
-                                                                        "if suffix/prefix are specified, template will be disabled\n"
-                                                                        "only commonly used templates are accepted:\n"
-                                                                        "https://github.com/ggerganov/llama.cpp/wiki/Templates-supported-by-llama_chat_apply_template" });
-    options.push_back({ "grammar" });
-    options.push_back({ "*",           "       --grammar GRAMMAR",      "BNF-like grammar to constrain generations (see samples in grammars/ dir) (default: '%s')", sparams.grammar.c_str() });
-    options.push_back({ "*",           "       --grammar-file FNAME",   "file to read grammar from" });
-    options.push_back({ "*",           "-j,    --json-schema SCHEMA",
-                                                                        "JSON schema to constrain generations (https://json-schema.org/), e.g. `{}` for any JSON object\n"
-                                                                        "For schemas w/ external $refs, use --grammar + example/json_schema_to_grammar.py instead" });
-
-    options.push_back({ "embedding" });
-    options.push_back({ "embedding",   "       --pooling {none,mean,cls,last}",
-                                                                        "pooling type for embeddings, use model default if unspecified" });
-    options.push_back({ "embedding",   "       --attention {causal,non-causal}",
-                                                                        "attention type for embeddings, use model default if unspecified" });
-
-    options.push_back({ "context hacking" });
-    options.push_back({ "*",           "       --rope-scaling {none,linear,yarn}",
-                                                                        "RoPE frequency scaling method, defaults to linear unless specified by the model" });
-    options.push_back({ "*",           "       --rope-scale N",         "RoPE context scaling factor, expands context by a factor of N" });
-    options.push_back({ "*",           "       --rope-freq-base N",     "RoPE base frequency, used by NTK-aware scaling (default: loaded from model)" });
-    options.push_back({ "*",           "       --rope-freq-scale N",    "RoPE frequency scaling factor, expands context by a factor of 1/N" });
-    options.push_back({ "*",           "       --yarn-orig-ctx N",      "YaRN: original context size of model (default: %d = model training context size)", params.yarn_orig_ctx });
-    options.push_back({ "*",           "       --yarn-ext-factor N",    "YaRN: extrapolation mix factor (default: %.1f, 0.0 = full interpolation)", (double)params.yarn_ext_factor });
-    options.push_back({ "*",           "       --yarn-attn-factor N",   "YaRN: scale sqrt(t) or attention magnitude (default: %.1f)", (double)params.yarn_attn_factor });
-    options.push_back({ "*",           "       --yarn-beta-slow N",     "YaRN: high correction dim or alpha (default: %.1f)", (double)params.yarn_beta_slow });
-    options.push_back({ "*",           "       --yarn-beta-fast N",     "YaRN: low correction dim or beta (default: %.1f)", (double)params.yarn_beta_fast });
-    options.push_back({ "*",           "-gan,  --grp-attn-n N",         "group-attention factor (default: %d)", params.grp_attn_n });
-    options.push_back({ "*",           "-gaw,  --grp-attn-w N",         "group-attention width (default: %.1f)", (double)params.grp_attn_w });
-    options.push_back({ "*",           "-dkvc, --dump-kv-cache",        "verbose print of the KV cache" });
-    options.push_back({ "*",           "-nkvo, --no-kv-offload",        "disable KV offload" });
-    options.push_back({ "*",           "-ctk,  --cache-type-k TYPE",    "KV cache data type for K (default: %s)", params.cache_type_k.c_str() });
-    options.push_back({ "*",           "-ctv,  --cache-type-v TYPE",    "KV cache data type for V (default: %s)", params.cache_type_v.c_str() });
-
-    options.push_back({ "perplexity" });
-    options.push_back({ "perplexity",  "       --all-logits",           "return logits for all tokens in the batch (default: %s)", params.logits_all ? "true" : "false" });
-    options.push_back({ "perplexity",  "       --hellaswag",            "compute HellaSwag score over random tasks from datafile supplied with -f" });
-    options.push_back({ "perplexity",  "       --hellaswag-tasks N",    "number of tasks to use when computing the HellaSwag score (default: %zu)", params.hellaswag_tasks });
-    options.push_back({ "perplexity",  "       --winogrande",           "compute Winogrande score over random tasks from datafile supplied with -f" });
-    options.push_back({ "perplexity",  "       --winogrande-tasks N",   "number of tasks to use when computing the Winogrande score (default: %zu)", params.winogrande_tasks });
-    options.push_back({ "perplexity",  "       --multiple-choice",      "compute multiple choice score over random tasks from datafile supplied with -f" });
-    options.push_back({ "perplexity",  "       --multiple-choice-tasks N",
-                                                                        "number of tasks to use when computing the multiple choice score (default: %zu)", params.multiple_choice_tasks });
-    options.push_back({ "perplexity",  "       --kl-divergence",        "computes KL-divergence to logits provided via --kl-divergence-base" });
-    options.push_back({ "perplexity",  "       --ppl-stride N",         "stride for perplexity calculation (default: %d)", params.ppl_stride });
-    options.push_back({ "perplexity",  "       --ppl-output-type {0,1}",
-                                                                        "output type for perplexity calculation (default: %d)", params.ppl_output_type });
-
-    options.push_back({ "parallel" });
-    options.push_back({ "*",           "-dt,   --defrag-thold N",       "KV cache defragmentation threshold (default: %.1f, < 0 - disabled)", (double)params.defrag_thold });
-    options.push_back({ "*",           "-np,   --parallel N",           "number of parallel sequences to decode (default: %d)", params.n_parallel });
-    options.push_back({ "*",           "-ns,   --sequences N",          "number of sequences to decode (default: %d)", params.n_sequences });
-    options.push_back({ "*",           "-cb,   --cont-batching",        "enable continuous batching (a.k.a dynamic batching) (default: %s)", params.cont_batching ? "enabled" : "disabled" });
-    options.push_back({ "*",           "-nocb, --no-cont-batching",     "disable continuous batching" });
-
-    options.push_back({ "multi-modality" });
-    options.push_back({ "*",           "       --mmproj FILE",          "path to a multimodal projector file for LLaVA. see examples/llava/README.md" });
-    options.push_back({ "*",           "       --image FILE",           "path to an image file. use with multimodal models. Specify multiple times for batching" });
-
-    options.push_back({ "backend" });
-    options.push_back({ "*",           "       --rpc SERVERS",          "comma separated list of RPC servers" });
-
-    if (llama_supports_mlock()) {
-        options.push_back({ "*",           "       --mlock",                "force system to keep model in RAM rather than swapping or compressing" });
-    }
-    if (llama_supports_mmap()) {
-        options.push_back({ "*",           "       --no-mmap",              "do not memory-map model (slower load but may reduce pageouts if not using mlock)" });
-    }
-    options.push_back({ "*",           "       --numa TYPE",            "attempt optimizations that help on some NUMA systems\n"
-                                                                        "  - distribute: spread execution evenly over all nodes\n"
-                                                                        "  - isolate: only spawn threads on CPUs on the node that execution started on\n"
-                                                                        "  - numactl: use the CPU map provided by numactl\n"
-                                                                        "if run without this previously, it is recommended to drop the system page cache before using this\n"
-                                                                        "see https://github.com/ggerganov/llama.cpp/issues/1437" });
-
-    if (llama_supports_gpu_offload()) {
-        options.push_back({ "*",           "-ngl,  --gpu-layers N",
-                                                                        "number of layers to store in VRAM" });
-        options.push_back({ "*",           "-ngld, --gpu-layers-draft N",
-                                                                        "number of layers to store in VRAM for the draft model" });
-        options.push_back({ "*",           "-sm,   --split-mode SPLIT_MODE",
-                                                                        "how to split the model across multiple GPUs, one of:\n"
-                                                                        "  - none: use one GPU only\n"
-                                                                        "  - layer (default): split layers and KV across GPUs\n"
-                                                                        "  - row: split rows across GPUs" });
-        options.push_back({ "*",           "-ts,   --tensor-split SPLIT",
-                                                                        "fraction of the model to offload to each GPU, comma-separated list of proportions, e.g. 3,1" });
-        options.push_back({ "*",           "-mg,   --main-gpu i",       "the GPU to use for the model (with split-mode = none),\n"
-                                                                        "or for intermediate results and KV (with split-mode = row) (default: %d)", params.main_gpu });
-    }
-
-    options.push_back({ "model" });
-    options.push_back({ "*",           "       --check-tensors",        "check model tensor data for invalid values (default: %s)", params.check_tensors ? "true" : "false" });
-    options.push_back({ "*",           "       --override-kv KEY=TYPE:VALUE",
-                                                                        "advanced option to override model metadata by key. may be specified multiple times.\n"
-                                                                        "types: int, float, bool, str. example: --override-kv tokenizer.ggml.add_bos_token=bool:false" });
-    options.push_back({ "*",           "       --lora FNAME",           "apply LoRA adapter (can be repeated to use multiple adapters)" });
-    options.push_back({ "*",           "       --lora-scaled FNAME S",  "apply LoRA adapter with user defined scaling S (can be repeated to use multiple adapters)" });
-    options.push_back({ "*",           "       --control-vector FNAME", "add a control vector\n"
-                                                                        "note: this argument can be repeated to add multiple control vectors" });
-    options.push_back({ "*",           "       --control-vector-scaled FNAME SCALE",
-                                                                        "add a control vector with user defined scaling SCALE\n"
-                                                                        "note: this argument can be repeated to add multiple scaled control vectors" });
-    options.push_back({ "*",           "       --control-vector-layer-range START END",
-                                                                        "layer range to apply the control vector(s) to, start and end inclusive" });
-    options.push_back({ "*",           "-m,    --model FNAME",          "model path (default: models/$filename with filename from --hf-file\n"
-                                                                        "or --model-url if set, otherwise %s)", DEFAULT_MODEL_PATH });
-    options.push_back({ "*",           "-md,   --model-draft FNAME",    "draft model for speculative decoding (default: unused)" });
-    options.push_back({ "*",           "-mu,   --model-url MODEL_URL",  "model download url (default: unused)" });
-    options.push_back({ "*",           "-hfr,  --hf-repo REPO",         "Hugging Face model repository (default: unused)" });
-    options.push_back({ "*",           "-hff,  --hf-file FILE",         "Hugging Face model file (default: unused)" });
-    options.push_back({ "*",           "-hft,  --hf-token TOKEN",       "Hugging Face access token (default: value from HF_TOKEN environment variable)" });
-
-    options.push_back({ "retrieval" });
-    options.push_back({ "retrieval",   "       --context-file FNAME",   "file to load context from (repeat to specify multiple files)" });
-    options.push_back({ "retrieval",   "       --chunk-size N",         "minimum length of embedded text chunks (default: %d)", params.chunk_size });
-    options.push_back({ "retrieval",   "       --chunk-separator STRING",
-                                                                        "separator between chunks (default: '%s')", params.chunk_separator.c_str() });
-
-    options.push_back({ "passkey" });
-    options.push_back({ "passkey",     "       --junk N",               "number of times to repeat the junk text (default: %d)", params.n_junk });
-    options.push_back({ "passkey",     "       --pos N",                "position of the passkey in the junk text (default: %d)", params.i_pos });
-
-    options.push_back({ "imatrix" });
-    options.push_back({ "imatrix",     "-o,    --output FNAME",         "output file (default: '%s')", params.out_file.c_str() });
-    options.push_back({ "imatrix",     "       --output-frequency N",   "output the imatrix every N iterations (default: %d)", params.n_out_freq });
-    options.push_back({ "imatrix",     "       --save-frequency N",     "save an imatrix copy every N iterations (default: %d)", params.n_save_freq });
-    options.push_back({ "imatrix",     "       --process-output",       "collect data for the output tensor (default: %s)", params.process_output ? "true" : "false" });
-    options.push_back({ "imatrix",     "       --no-ppl",               "do not compute perplexity (default: %s)", params.compute_ppl ? "true" : "false" });
-    options.push_back({ "imatrix",     "       --chunk N",              "start processing the input from chunk N (default: %d)", params.i_chunk });
-
-    options.push_back({ "bench" });
-    options.push_back({ "bench",       "-pps",                          "is the prompt shared across parallel sequences (default: %s)", params.is_pp_shared ? "true" : "false" });
-    options.push_back({ "bench",       "-npp n0,n1,...",                "number of prompt tokens" });
-    options.push_back({ "bench",       "-ntg n0,n1,...",                "number of text generation tokens" });
-    options.push_back({ "bench",       "-npl n0,n1,...",                "number of parallel prompts" });
-
-    options.push_back({ "embedding" });
-    options.push_back({ "embedding",   "       --embd-normalize",       "normalisation for embendings (default: %d) (-1=none, 0=max absolute int16, 1=taxicab, 2=euclidean, >2=p-norm)", params.embd_normalize });
-    options.push_back({ "embedding",   "       --embd-output-format",   "empty = default, \"array\" = [[],[]...], \"json\" = openai style, \"json+\" = same \"json\" + cosine similarity matrix" });
-    options.push_back({ "embedding",   "       --embd-separator",       "separator of embendings (default \\n) for example \"<#sep#>\"" });
-
-    options.push_back({ "server" });
-    options.push_back({ "server",      "       --host HOST",            "ip address to listen (default: %s)", params.hostname.c_str() });
-    options.push_back({ "server",      "       --port PORT",            "port to listen (default: %d)", params.port });
-    options.push_back({ "server",      "       --path PATH",            "path to serve static files from (default: %s)", params.public_path.c_str() });
-    options.push_back({ "server",      "       --embedding(s)",         "enable embedding endpoint (default: %s)", params.embedding ? "enabled" : "disabled" });
-    options.push_back({ "server",      "       --api-key KEY",          "API key to use for authentication (default: none)" });
-    options.push_back({ "server",      "       --api-key-file FNAME",   "path to file containing API keys (default: none)" });
-    options.push_back({ "server",      "       --ssl-key-file FNAME",   "path to file a PEM-encoded SSL private key" });
-    options.push_back({ "server",      "       --ssl-cert-file FNAME",  "path to file a PEM-encoded SSL certificate" });
-    options.push_back({ "server",      "       --timeout N",            "server read/write timeout in seconds (default: %d)", params.timeout_read });
-    options.push_back({ "server",      "       --threads-http N",       "number of threads used to process HTTP requests (default: %d)", params.n_threads_http });
-    options.push_back({ "server",      "       --system-prompt-file FNAME",
-                                                                        "set a file to load a system prompt (initial prompt of all slots), this is useful for chat applications" });
-    options.push_back({ "server",      "       --log-format {text,json}",
-                                                                        "log output format: json or text (default: json)" });
-    options.push_back({ "server",      "       --metrics",              "enable prometheus compatible metrics endpoint (default: %s)", params.endpoint_metrics ? "enabled" : "disabled" });
-    options.push_back({ "server",      "       --no-slots",             "disables slots monitoring endpoint (default: %s)", params.endpoint_slots ? "enabled" : "disabled" });
-    options.push_back({ "server",      "       --slot-save-path PATH",  "path to save slot kv cache (default: disabled)" });
-    options.push_back({ "server",      "       --chat-template JINJA_TEMPLATE",
-                                                                        "set custom jinja chat template (default: template taken from model's metadata)\n"
-                                                                        "only commonly used templates are accepted:\n"
-                                                                        "https://github.com/ggerganov/llama.cpp/wiki/Templates-supported-by-llama_chat_apply_template" });
-    options.push_back({ "server",      "-sps,  --slot-prompt-similarity SIMILARITY",
-                                                                        "how much the prompt of a request must match the prompt of a slot in order to use that slot (default: %.2f, 0.0 = disabled)\n", params.slot_prompt_similarity });
-
-#ifndef LOG_DISABLE_LOGS
-    options.push_back({ "logging" });
-    options.push_back({ "*",           "       --simple-io",            "use basic IO for better compatibility in subprocesses and limited consoles" });
-    options.push_back({ "*",           "-ld,   --logdir LOGDIR",        "path under which to save YAML logs (no logging if unset)" });
-    options.push_back({ "logging",     "       --log-test",             "Run simple logging test" });
-    options.push_back({ "logging",     "       --log-disable",          "Disable trace logs" });
-    options.push_back({ "logging",     "       --log-enable",           "Enable trace logs" });
-    options.push_back({ "logging",     "       --log-file FNAME",       "Specify a log filename (without extension)" });
-    options.push_back({ "logging",     "       --log-new",              "Create a separate new log file on start. "
-                                                                        "Each log file will have unique name: \"<name>.<ID>.log\"" });
-    options.push_back({ "logging",     "       --log-append",           "Don't truncate the old log file." });
-#endif // LOG_DISABLE_LOGS
-
-    options.push_back({ "cvector" });
-    options.push_back({ "cvector",     "-o,    --output FNAME",         "output file (default: '%s')", params.cvector_outfile.c_str() });
-    options.push_back({ "cvector",     "       --positive-file FNAME",  "positive prompts file, one prompt per line (default: '%s')", params.cvector_positive_file.c_str() });
-    options.push_back({ "cvector",     "       --negative-file FNAME",  "negative prompts file, one prompt per line (default: '%s')", params.cvector_negative_file.c_str() });
-    options.push_back({ "cvector",     "       --pca-batch N",          "batch size used for PCA. Larger batch runs faster, but uses more memory (default: %d)", params.n_pca_batch });
-    options.push_back({ "cvector",     "       --pca-iter N",           "number of iterations used for PCA (default: %d)", params.n_pca_iterations });
-    options.push_back({ "cvector",     "       --method {pca,mean}",    "dimensionality reduction method to be used (default: pca)" });
-
-    options.push_back({ "export-lora" });
-    options.push_back({ "export-lora", "-m,    --model",                "model path from which to load base model (default '%s')", params.model.c_str() });
-    options.push_back({ "export-lora", "       --lora FNAME",           "path to LoRA adapter  (can be repeated to use multiple adapters)" });
-    options.push_back({ "export-lora", "       --lora-scaled FNAME S",  "path to LoRA adapter with user defined scaling S  (can be repeated to use multiple adapters)" });
-    options.push_back({ "*",           "-t,    --threads N",            "number of threads to use during computation (default: %d)", params.n_threads });
-    options.push_back({ "export-lora", "-o,    --output FNAME",         "output file (default: '%s')", params.lora_outfile.c_str() });
-
-    printf("usage: %s [options]\n", argv[0]);
-
-    for (const auto & o : options) {
-        if (!o.grp.empty()) {
-            printf("\n%s:\n\n", o.grp.c_str());
-            continue;
-        }
-        printf("  %-32s", o.args.c_str());
-        if (o.args.length() > 30) {
-            printf("\n%34s", "");
-        }
-
-        const auto desc = o.desc;
-        size_t start = 0;
-        size_t end = desc.find('\n');
-        while (end != std::string::npos) {
-            printf("%s\n%34s", desc.substr(start, end - start).c_str(), "");
-            start = end + 1;
-            end = desc.find('\n', start);
-        }
-
-        printf("%s\n", desc.substr(start).c_str());
-    }
-    printf("\n");
+#ifdef NDEBUG
+    const char * build_type = "";
+#else
+    const char * build_type = " (debug)";
+#endif
+
+    LOG_INF("build: %d (%s) with %s for %s%s\n", LLAMA_BUILD_NUMBER, LLAMA_COMMIT, LLAMA_COMPILER, LLAMA_BUILD_TARGET, build_type);
 }
 
 std::string gpt_params_get_system_info(const gpt_params & params) {
     std::ostringstream os;
 
-    os << "system_info: n_threads = " << params.n_threads;
-    if (params.n_threads_batch != -1) {
-        os << " (n_threads_batch = " << params.n_threads_batch << ")";
+    os << "system_info: n_threads = " << params.cpuparams.n_threads;
+    if (params.cpuparams_batch.n_threads != -1) {
+        os << " (n_threads_batch = " << params.cpuparams_batch.n_threads << ")";
     }
+#if defined(_WIN32) && (_WIN32_WINNT >= 0x0601) && !defined(__MINGW64__) // windows 7 and later
+    // TODO: windows + arm64 + mingw64
+    DWORD logicalProcessorCount = GetActiveProcessorCount(ALL_PROCESSOR_GROUPS);
+    os << " / " << logicalProcessorCount << " | " << llama_print_system_info();
+#else
     os << " / " << std::thread::hardware_concurrency() << " | " << llama_print_system_info();
+#endif
 
     return os.str();
 }
@@ -1766,6 +441,111 @@ std::string string_get_sortable_timestamp() {
     return std::string(timestamp_no_ns) + "." + std::string(timestamp_ns);
 }
 
+void string_replace_all(std::string & s, const std::string & search, const std::string & replace) {
+    if (search.empty()) {
+        return;
+    }
+    std::string builder;
+    builder.reserve(s.length());
+    size_t pos = 0;
+    size_t last_pos = 0;
+    while ((pos = s.find(search, last_pos)) != std::string::npos) {
+        builder.append(s, last_pos, pos - last_pos);
+        builder.append(replace);
+        last_pos = pos + search.length();
+    }
+    builder.append(s, last_pos, std::string::npos);
+    s = std::move(builder);
+}
+
+std::string string_from(bool value) {
+    return value ? "true" : "false";
+}
+
+std::string string_from(const std::vector<int> & values) {
+    std::stringstream buf;
+
+    buf << "[ ";
+    bool first = true;
+    for (auto e : values) {
+        if (first) {
+            first = false;
+        } else {
+            buf << ", ";
+        }
+        buf << std::to_string(e);
+    }
+    buf << " ]";
+
+    return buf.str();
+}
+
+std::string string_from(const struct llama_context * ctx, const std::vector<llama_token> & tokens) {
+    std::stringstream buf;
+
+    buf << "[ ";
+
+    bool first = true;
+    for (const auto & token : tokens) {
+        if (!first) {
+            buf << ", ";
+        } else {
+            first = false;
+        }
+
+        auto detokenized = llama_token_to_piece(ctx, token);
+
+        detokenized.erase(
+            std::remove_if(
+                detokenized.begin(),
+                detokenized.end(),
+                [](const unsigned char c) { return !std::isprint(c); }),
+            detokenized.end());
+
+        buf << "'" << detokenized << "'"
+            << ":" << std::to_string(token);
+    }
+
+    buf << " ]";
+
+    return buf.str();
+}
+
+std::string string_from(const struct llama_context * ctx, const struct llama_batch & batch) {
+    std::stringstream buf;
+
+    buf << "[ ";
+
+    bool first = true;
+    for (int i = 0; i < batch.n_tokens; ++i) {
+        if (!first) {
+            buf << ", ";
+        } else {
+            first = false;
+        }
+
+        auto detokenized = llama_token_to_piece(ctx, batch.token[i]);
+
+        detokenized.erase(
+                std::remove_if(
+                    detokenized.begin(),
+                    detokenized.end(),
+                    [](const unsigned char c) { return !std::isprint(c); }),
+                detokenized.end());
+
+        buf << "\n" << std::to_string(i)
+            << ":token '" << detokenized << "'"
+            << ":pos " << std::to_string(batch.pos[i])
+            << ":n_seq_id  " << std::to_string(batch.n_seq_id[i])
+            << ":seq_id " << std::to_string(batch.seq_id[i][0])
+            << ":logits " << std::to_string(batch.logits[i]);
+    }
+
+    buf << " ]";
+
+    return buf.str();
+}
+
 void string_process_escapes(std::string & input) {
     std::size_t input_len = input.length();
     std::size_t output_idx = 0;
@@ -1806,7 +586,7 @@ void string_process_escapes(std::string & input) {
 bool string_parse_kv_override(const char * data, std::vector<llama_model_kv_override> & overrides) {
     const char * sep = strchr(data, '=');
     if (sep == nullptr || sep - data >= 128) {
-        fprintf(stderr, "%s: malformed KV override '%s'\n", __func__, data);
+        LOG_ERR("%s: malformed KV override '%s'\n", __func__, data);
         return false;
     }
     llama_model_kv_override kvo;
@@ -1829,20 +609,20 @@ bool string_parse_kv_override(const char * data, std::vector<llama_model_kv_over
         } else if (std::strcmp(sep, "false") == 0) {
             kvo.val_bool = false;
         } else {
-            fprintf(stderr, "%s: invalid boolean value for KV override '%s'\n", __func__, data);
+            LOG_ERR("%s: invalid boolean value for KV override '%s'\n", __func__, data);
             return false;
         }
     } else if (strncmp(sep, "str:", 4) == 0) {
         sep += 4;
         kvo.tag = LLAMA_KV_OVERRIDE_TYPE_STR;
         if (strlen(sep) > 127) {
-            fprintf(stderr, "%s: malformed KV override '%s', value cannot exceed 127 chars\n", __func__, data);
+            LOG_ERR("%s: malformed KV override '%s', value cannot exceed 127 chars\n", __func__, data);
             return false;
         }
         strncpy(kvo.val_str, sep, 127);
         kvo.val_str[127] = '\0';
     } else {
-        fprintf(stderr, "%s: invalid type for KV override '%s'\n", __func__, data);
+        LOG_ERR("%s: invalid type for KV override '%s'\n", __func__, data);
         return false;
     }
     overrides.emplace_back(std::move(kvo));
@@ -2039,8 +819,8 @@ std::string fs_get_cache_file(const std::string & filename) {
 //
 // Model utils
 //
-
-std::tuple<struct llama_model *, struct llama_context *> llama_init_from_gpt_params(gpt_params & params) {
+struct llama_init_result llama_init_from_gpt_params(gpt_params & params) {
+    llama_init_result iparams;
     auto mparams = llama_model_params_from_gpt_params(params);
 
     llama_model * model = nullptr;
@@ -2054,17 +834,42 @@ std::tuple<struct llama_model *, struct llama_context *> llama_init_from_gpt_par
     }
 
     if (model == NULL) {
-        fprintf(stderr, "%s: error: failed to load model '%s'\n", __func__, params.model.c_str());
-        return std::make_tuple(nullptr, nullptr);
+        LOG_ERR("%s: failed to load model '%s'\n", __func__, params.model.c_str());
+        return iparams;
+    }
+
+    if (params.reranking) {
+        bool ok = true;
+
+        if (llama_token_bos(model) == LLAMA_TOKEN_NULL) {
+            LOG_WRN("%s: warning: model does not have a  BOS token, reranking will not work\n", __func__);
+            ok = false;
+        }
+
+        if (llama_token_eos(model) == LLAMA_TOKEN_NULL) {
+            LOG_WRN("%s: warning: model does not have an EOS token, reranking will not work\n", __func__);
+            ok = false;
+        }
+
+        if (llama_token_sep(model) == LLAMA_TOKEN_NULL) {
+            LOG_WRN("%s: warning: model does not have a  SEP token, reranking will not work\n", __func__);
+            ok = false;
+        }
+
+        if (!ok) {
+            llama_free_model(model);
+
+            return iparams;
+        }
     }
 
     auto cparams = llama_context_params_from_gpt_params(params);
 
     llama_context * lctx = llama_new_context_with_model(model, cparams);
     if (lctx == NULL) {
-        fprintf(stderr, "%s: error: failed to create context with model '%s'\n", __func__, params.model.c_str());
+        LOG_ERR("%s: failed to create context with model '%s'\n", __func__, params.model.c_str());
         llama_free_model(model);
-        return std::make_tuple(nullptr, nullptr);
+        return iparams;
     }
 
     if (!params.control_vectors.empty()) {
@@ -2075,7 +880,8 @@ std::tuple<struct llama_model *, struct llama_context *> llama_init_from_gpt_par
         if (cvec.n_embd == -1) {
             llama_free(lctx);
             llama_free_model(model);
-            return std::make_tuple(nullptr, nullptr);
+
+            return iparams;
         }
 
         int err = llama_control_vector_apply(lctx,
@@ -2087,38 +893,50 @@ std::tuple<struct llama_model *, struct llama_context *> llama_init_from_gpt_par
         if (err) {
             llama_free(lctx);
             llama_free_model(model);
-            return std::make_tuple(nullptr, nullptr);
+
+            return iparams;
         }
     }
 
-    for (unsigned int i = 0; i < params.lora_adapter.size(); ++i) {
-        const std::string & lora_adapter = std::get<0>(params.lora_adapter[i]);
-        float lora_scale = std::get<1>(params.lora_adapter[i]);
-        auto adapter = llama_lora_adapter_init(model, lora_adapter.c_str());
-        if (adapter == nullptr) {
-            fprintf(stderr, "%s: error: failed to apply lora adapter\n", __func__);
+    // load and optionally apply lora adapters
+    for (auto & la : params.lora_adapters) {
+        llama_lora_adapter_container loaded_la;
+        loaded_la.path = la.path;
+        loaded_la.scale = la.scale;
+        loaded_la.adapter = llama_lora_adapter_init(model, la.path.c_str());
+        if (loaded_la.adapter == nullptr) {
+            LOG_ERR("%s: failed to apply lora adapter '%s'\n", __func__, la.path.c_str());
             llama_free(lctx);
             llama_free_model(model);
-            return std::make_tuple(nullptr, nullptr);
+            return iparams;
         }
-        llama_lora_adapter_set(lctx, adapter, lora_scale);
+        iparams.lora_adapters.push_back(loaded_la); // copy to list of loaded adapters
+    }
+    if (!params.lora_init_without_apply) {
+        llama_lora_adapters_apply(lctx, iparams.lora_adapters);
     }
 
-    if (params.ignore_eos) {
-        params.sparams.logit_bias[llama_token_eos(model)] = -INFINITY;
+    if (params.sparams.ignore_eos && llama_token_eos(model) == LLAMA_TOKEN_NULL) {
+        LOG_WRN("%s: warning: model does not have an EOS token, ignoring --ignore-eos\n", __func__);
+        params.sparams.ignore_eos = false;
     }
 
     if (params.warmup) {
-        LOG("warming up the model with an empty run\n");
+        LOG_WRN("%s: warming up the model with an empty run - please wait ... (--no-warmup to disable)\n", __func__);
 
         std::vector<llama_token> tmp;
         llama_token bos = llama_token_bos(model);
         llama_token eos = llama_token_eos(model);
         // some models (e.g. T5) don't have a BOS token
-        if (bos != -1) {
+        if (bos != LLAMA_TOKEN_NULL) {
             tmp.push_back(bos);
         }
-        tmp.push_back(eos);
+        if (eos != LLAMA_TOKEN_NULL) {
+            tmp.push_back(eos);
+        }
+        if (tmp.empty()) {
+            tmp.push_back(0);
+        }
 
         if (llama_model_has_encoder(model)) {
             llama_encode(lctx, llama_batch_get_one(tmp.data(), tmp.size(), 0, 0));
@@ -2129,13 +947,27 @@ std::tuple<struct llama_model *, struct llama_context *> llama_init_from_gpt_par
             tmp.clear();
             tmp.push_back(decoder_start_token_id);
         }
-        llama_decode(lctx, llama_batch_get_one(tmp.data(), std::min(tmp.size(), (size_t) params.n_batch), 0, 0));
+        if (llama_model_has_decoder(model)) {
+            llama_decode(lctx, llama_batch_get_one(tmp.data(), std::min(tmp.size(), (size_t) params.n_batch), 0, 0));
+        }
         llama_kv_cache_clear(lctx);
         llama_synchronize(lctx);
-        llama_reset_timings(lctx);
+        llama_perf_context_reset(lctx);
     }
 
-    return std::make_tuple(model, lctx);
+    iparams.model   = model;
+    iparams.context = lctx;
+
+    return iparams;
+}
+
+void llama_lora_adapters_apply(struct llama_context * ctx, std::vector<llama_lora_adapter_container> & lora_adapters) {
+    llama_lora_adapter_clear(ctx);
+    for (auto & la : lora_adapters) {
+        if (la.scale != 0.0f) {
+            llama_lora_adapter_set(ctx, la.adapter, la.scale);
+        }
+    }
 }
 
 struct llama_model_params llama_model_params_from_gpt_params(const gpt_params & params) {
@@ -2197,9 +1029,9 @@ struct llama_context_params llama_context_params_from_gpt_params(const gpt_param
     cparams.n_seq_max         = params.n_parallel;
     cparams.n_batch           = params.n_batch;
     cparams.n_ubatch          = params.n_ubatch;
-    cparams.n_threads         = params.n_threads;
-    cparams.n_threads_batch   = params.n_threads_batch == -1 ? params.n_threads : params.n_threads_batch;
-    cparams.seed              = params.seed;
+    cparams.n_threads         = params.cpuparams.n_threads;
+    cparams.n_threads_batch   = params.cpuparams_batch.n_threads == -1 ?
+                                    params.cpuparams.n_threads : params.cpuparams_batch.n_threads;
     cparams.logits_all        = params.logits_all;
     cparams.embeddings        = params.embedding;
     cparams.rope_scaling_type = params.rope_scaling_type;
@@ -2217,6 +1049,12 @@ struct llama_context_params llama_context_params_from_gpt_params(const gpt_param
     cparams.cb_eval_user_data = params.cb_eval_user_data;
     cparams.offload_kqv       = !params.no_kv_offload;
     cparams.flash_attn        = params.flash_attn;
+    cparams.no_perf           = params.no_perf;
+
+    if (params.reranking) {
+        cparams.embeddings    = true;
+        cparams.pooling_type  = LLAMA_POOLING_TYPE_RANK;
+    }
 
     cparams.type_k = kv_cache_type_from_str(params.cache_type_k);
     cparams.type_v = kv_cache_type_from_str(params.cache_type_v);
@@ -2224,19 +1062,62 @@ struct llama_context_params llama_context_params_from_gpt_params(const gpt_param
     return cparams;
 }
 
+struct ggml_threadpool_params ggml_threadpool_params_from_cpu_params(const cpu_params & params) {
+    struct ggml_threadpool_params tpp;
+
+    ggml_threadpool_params_init(&tpp, params.n_threads); // setup the defaults
+
+    if (params.mask_valid) {
+        std::memcpy(&tpp.cpumask, &params.cpumask, GGML_MAX_N_THREADS);
+    }
+
+    tpp.prio       = params.priority;
+    tpp.poll       = params.poll;
+    tpp.strict_cpu = params.strict_cpu;
+
+    return tpp;
+}
+
 #ifdef LLAMA_USE_CURL
 
+#define CURL_MAX_RETRY 3
+#define CURL_RETRY_DELAY_SECONDS 2
+
+
 static bool starts_with(const std::string & str, const std::string & prefix) {
     // While we wait for C++20's std::string::starts_with...
     return str.rfind(prefix, 0) == 0;
 }
 
+static bool curl_perform_with_retry(const std::string& url, CURL* curl, int max_attempts, int retry_delay_seconds) {
+    int remaining_attempts = max_attempts;
+
+    while (remaining_attempts > 0) {
+        LOG_INF("%s: Trying to download from %s (attempt %d of %d)...\n", __func__ , url.c_str(), max_attempts - remaining_attempts + 1, max_attempts);
+
+        CURLcode res = curl_easy_perform(curl);
+        if (res == CURLE_OK) {
+            return true;
+        }
+
+        int exponential_backoff_delay = std::pow(retry_delay_seconds, max_attempts - remaining_attempts) * 1000;
+        LOG_WRN("%s: curl_easy_perform() failed: %s, retrying after %d milliseconds...\n", __func__, curl_easy_strerror(res), exponential_backoff_delay);
+
+        remaining_attempts--;
+        std::this_thread::sleep_for(std::chrono::milliseconds(exponential_backoff_delay));
+    }
+
+    LOG_ERR("%s: curl_easy_perform() failed after %d attempts\n", __func__, max_attempts);
+
+    return false;
+}
+
 static bool llama_download_file(const std::string & url, const std::string & path, const std::string & hf_token) {
 
     // Initialize libcurl
     std::unique_ptr<CURL, decltype(&curl_easy_cleanup)> curl(curl_easy_init(), &curl_easy_cleanup);
     if (!curl) {
-        fprintf(stderr, "%s: error initializing libcurl\n", __func__);
+        LOG_ERR("%s: error initializing libcurl\n", __func__);
         return false;
     }
 
@@ -2277,11 +1158,11 @@ static bool llama_download_file(const std::string & url, const std::string & pat
         if (metadata_in.good()) {
             try {
                 metadata_in >> metadata;
-                fprintf(stderr, "%s: previous metadata file found %s: %s\n", __func__, metadata_path.c_str(), metadata.dump().c_str());
+                LOG_INF("%s: previous metadata file found %s: %s\n", __func__, metadata_path.c_str(), metadata.dump().c_str());
                 if (metadata.contains("url") && metadata.at("url").is_string()) {
                     auto previous_url = metadata.at("url").get<std::string>();
                     if (previous_url != url) {
-                        fprintf(stderr, "%s: Model URL mismatch: %s != %s\n", __func__, url.c_str(), previous_url.c_str());
+                        LOG_ERR("%s: Model URL mismatch: %s != %s\n", __func__, url.c_str(), previous_url.c_str());
                         return false;
                     }
                 }
@@ -2292,12 +1173,12 @@ static bool llama_download_file(const std::string & url, const std::string & pat
                     last_modified = metadata.at("lastModified");
                 }
             } catch (const nlohmann::json::exception & e) {
-                fprintf(stderr, "%s: error reading metadata file %s: %s\n", __func__, metadata_path.c_str(), e.what());
+            LOG_ERR("%s: error reading metadata file %s: %s\n", __func__, metadata_path.c_str(), e.what());
                 return false;
             }
         }
     } else {
-        fprintf(stderr, "%s: no previous model file found %s\n", __func__, path.c_str());
+        LOG_INF("%s: no previous model file found %s\n", __func__, path.c_str());
     }
 
     // Send a HEAD request to retrieve the etag and last-modified headers
@@ -2334,9 +1215,8 @@ static bool llama_download_file(const std::string & url, const std::string & pat
         curl_easy_setopt(curl.get(), CURLOPT_HEADERFUNCTION, static_cast<CURLOPT_HEADERFUNCTION_PTR>(header_callback));
         curl_easy_setopt(curl.get(), CURLOPT_HEADERDATA, &headers);
 
-        CURLcode res = curl_easy_perform(curl.get());
-        if (res != CURLE_OK) {
-            fprintf(stderr, "%s: curl_easy_perform() failed: %s\n", __func__, curl_easy_strerror(res));
+        bool was_perform_successful = curl_perform_with_retry(url, curl.get(), CURL_MAX_RETRY, CURL_RETRY_DELAY_SECONDS);
+        if (!was_perform_successful) {
             return false;
         }
 
@@ -2346,26 +1226,26 @@ static bool llama_download_file(const std::string & url, const std::string & pat
             // HEAD not supported, we don't know if the file has changed
             // force trigger downloading
             force_download = true;
-            fprintf(stderr, "%s: HEAD invalid http status code received: %ld\n", __func__, http_code);
+            LOG_ERR("%s: HEAD invalid http status code received: %ld\n", __func__, http_code);
         }
     }
 
     bool should_download = !file_exists || force_download;
     if (!should_download) {
         if (!etag.empty() && etag != headers.etag) {
-            fprintf(stderr, "%s: ETag header is different (%s != %s): triggering a new download\n", __func__, etag.c_str(), headers.etag.c_str());
+            LOG_WRN("%s: ETag header is different (%s != %s): triggering a new download\n", __func__, etag.c_str(), headers.etag.c_str());
             should_download = true;
         } else if (!last_modified.empty() && last_modified != headers.last_modified) {
-            fprintf(stderr, "%s: Last-Modified header is different (%s != %s): triggering a new download\n", __func__, last_modified.c_str(), headers.last_modified.c_str());
+            LOG_WRN("%s: Last-Modified header is different (%s != %s): triggering a new download\n", __func__, last_modified.c_str(), headers.last_modified.c_str());
             should_download = true;
         }
     }
     if (should_download) {
         std::string path_temporary = path + ".downloadInProgress";
         if (file_exists) {
-            fprintf(stderr, "%s: deleting previous downloaded file: %s\n", __func__, path.c_str());
+            LOG_WRN("%s: deleting previous downloaded file: %s\n", __func__, path.c_str());
             if (remove(path.c_str()) != 0) {
-                fprintf(stderr, "%s: unable to delete file: %s\n", __func__, path.c_str());
+                LOG_ERR("%s: unable to delete file: %s\n", __func__, path.c_str());
                 return false;
             }
         }
@@ -2380,7 +1260,7 @@ static bool llama_download_file(const std::string & url, const std::string & pat
 
         std::unique_ptr<FILE, FILE_deleter> outfile(fopen(path_temporary.c_str(), "wb"));
         if (!outfile) {
-            fprintf(stderr, "%s: error opening local file for writing: %s\n", __func__, path.c_str());
+            LOG_ERR("%s: error opening local file for writing: %s\n", __func__, path.c_str());
             return false;
         }
 
@@ -2411,18 +1291,17 @@ static bool llama_download_file(const std::string & url, const std::string & pat
         };
 
         // start the download
-        fprintf(stderr, "%s: downloading from %s to %s (server_etag:%s, server_last_modified:%s)...\n", __func__,
-                llama_download_hide_password_in_url(url).c_str(), path.c_str(), headers.etag.c_str(), headers.last_modified.c_str());
-        auto res = curl_easy_perform(curl.get());
-        if (res != CURLE_OK) {
-            fprintf(stderr, "%s: curl_easy_perform() failed: %s\n", __func__, curl_easy_strerror(res));
+        LOG_INF("%s: trying to download model from %s to %s (server_etag:%s, server_last_modified:%s)...\n", __func__,
+            llama_download_hide_password_in_url(url).c_str(), path.c_str(), headers.etag.c_str(), headers.last_modified.c_str());
+        bool was_perform_successful = curl_perform_with_retry(url, curl.get(), CURL_MAX_RETRY, CURL_RETRY_DELAY_SECONDS);
+        if (!was_perform_successful) {
             return false;
         }
 
         long http_code = 0;
         curl_easy_getinfo (curl.get(), CURLINFO_RESPONSE_CODE, &http_code);
         if (http_code < 200 || http_code >= 400) {
-            fprintf(stderr, "%s: invalid http status code received: %ld\n", __func__, http_code);
+            LOG_ERR("%s: invalid http status code received: %ld\n", __func__, http_code);
             return false;
         }
 
@@ -2436,10 +1315,10 @@ static bool llama_download_file(const std::string & url, const std::string & pat
             {"lastModified", headers.last_modified}
         });
         std::ofstream(metadata_path) << metadata.dump(4);
-        fprintf(stderr, "%s: file metadata saved: %s\n", __func__, metadata_path.c_str());
+        LOG_INF("%s: file metadata saved: %s\n", __func__, metadata_path.c_str());
 
         if (rename(path_temporary.c_str(), path.c_str()) != 0) {
-            fprintf(stderr, "%s: unable to rename file: %s to %s\n", __func__, path_temporary.c_str(), path.c_str());
+            LOG_ERR("%s: unable to rename file: %s to %s\n", __func__, path_temporary.c_str(), path.c_str());
             return false;
         }
     }
@@ -2454,7 +1333,7 @@ struct llama_model * llama_load_model_from_url(
         const struct llama_model_params & params) {
     // Basic validation of the model_url
     if (!model_url || strlen(model_url) == 0) {
-        fprintf(stderr, "%s: invalid model_url\n", __func__);
+        LOG_ERR("%s: invalid model_url\n", __func__);
         return NULL;
     }
 
@@ -2471,7 +1350,7 @@ struct llama_model * llama_load_model_from_url(
         };
         auto * ctx_gguf = gguf_init_from_file(path_model, gguf_params);
         if (!ctx_gguf) {
-            fprintf(stderr, "\n%s:  failed to load input GGUF from %s\n", __func__, path_model);
+            LOG_ERR("\n%s:  failed to load input GGUF from %s\n", __func__, path_model);
             return NULL;
         }
 
@@ -2491,14 +1370,12 @@ struct llama_model * llama_load_model_from_url(
         // and extract split URL and PATH prefixes
         {
             if (!llama_split_prefix(split_prefix, sizeof(split_prefix), path_model, 0, n_split)) {
-                fprintf(stderr, "\n%s: unexpected model file name: %s"
-                                " n_split=%d\n", __func__, path_model, n_split);
+                LOG_ERR("\n%s: unexpected model file name: %s n_split=%d\n", __func__, path_model, n_split);
                 return NULL;
             }
 
             if (!llama_split_prefix(split_url_prefix, sizeof(split_url_prefix), model_url, 0, n_split)) {
-                fprintf(stderr, "\n%s: unexpected model url: %s"
-                                " n_split=%d\n", __func__, model_url, n_split);
+                LOG_ERR("\n%s: unexpected model url: %s n_split=%d\n", __func__, model_url, n_split);
                 return NULL;
             }
         }
@@ -2558,7 +1435,7 @@ struct llama_model * llama_load_model_from_url(
         const char * /*path_model*/,
         const char * /*hf_token*/,
         const struct llama_model_params & /*params*/) {
-    fprintf(stderr, "%s: llama.cpp built without libcurl, downloading from an url not supported.\n", __func__);
+    LOG_WRN("%s: llama.cpp built without libcurl, downloading from an url not supported.\n", __func__);
     return nullptr;
 }
 
@@ -2568,7 +1445,7 @@ struct llama_model * llama_load_model_from_hf(
         const char * /*path_model*/,
         const char * /*hf_token*/,
         const struct llama_model_params & /*params*/) {
-    fprintf(stderr, "%s: llama.cpp built without libcurl, downloading from Hugging Face not supported.\n", __func__);
+    LOG_WRN("%s: llama.cpp built without libcurl, downloading from Hugging Face not supported.\n", __func__);
     return nullptr;
 }
 
@@ -2588,6 +1465,8 @@ void llama_batch_add(
                           llama_pos   pos,
     const std::vector<llama_seq_id> & seq_ids,
                                bool   logits) {
+    GGML_ASSERT(batch.seq_id[batch.n_tokens] && "llama_batch size exceeded");
+
     batch.token   [batch.n_tokens] = id;
     batch.pos     [batch.n_tokens] = pos;
     batch.n_seq_id[batch.n_tokens] = seq_ids.size();
@@ -2662,12 +1541,6 @@ std::string llama_detokenize(llama_context * ctx, const std::vector<llama_token>
     return text;
 }
 
-bool llama_should_add_bos_token(const llama_model * model) {
-    const int add_bos = llama_add_bos_token(model);
-
-    return add_bos != -1 ? bool(add_bos) : (llama_vocab_type(model) == LLAMA_VOCAB_TYPE_SPM);
-}
-
 //
 // Chat template utils
 //
@@ -2902,13 +1775,13 @@ static llama_control_vector_data llama_control_vector_load_one(const llama_contr
     };
     struct gguf_context * ctx_gguf = gguf_init_from_file(load_info.fname.c_str(), meta_gguf_params);
     if (!ctx_gguf) {
-        fprintf(stderr, "%s: failed to load control vector file from %s\n", __func__, load_info.fname.c_str());
+        LOG_ERR("%s: failed to load control vector file from %s\n", __func__, load_info.fname.c_str());
         return result;
     }
 
     int32_t n_tensors = gguf_get_n_tensors(ctx_gguf);
     if (n_tensors == 0) {
-        fprintf(stderr, "%s: no direction tensors found in %s\n", __func__, load_info.fname.c_str());
+        LOG_WRN("%s: no direction tensors found in %s\n", __func__, load_info.fname.c_str());
     }
 
     for (int i = 0; i < n_tensors; i++) {
@@ -2926,23 +1799,23 @@ static llama_control_vector_data llama_control_vector_load_one(const llama_contr
             }
         }
         if (layer_idx < 0) {
-            fprintf(stderr, "%s: invalid/unparsable direction tensor layer index in %s\n", __func__, load_info.fname.c_str());
+            LOG_ERR("%s: invalid/unparsable direction tensor layer index in %s\n", __func__, load_info.fname.c_str());
             result.n_embd = -1;
             break;
         } else if (layer_idx == 0) {
-            fprintf(stderr, "%s: invalid (zero) direction tensor layer index in %s\n", __func__, load_info.fname.c_str());
+            LOG_ERR("%s: invalid (zero) direction tensor layer index in %s\n", __func__, load_info.fname.c_str());
             result.n_embd = -1;
             break;
         }
 
         struct ggml_tensor * tensor = ggml_get_tensor(ctx, name.c_str());
         if (tensor->type != GGML_TYPE_F32) {
-            fprintf(stderr, "%s: invalid (non-F32) direction tensor type in %s\n", __func__, load_info.fname.c_str());
+            LOG_ERR("%s: invalid (non-F32) direction tensor type in %s\n", __func__, load_info.fname.c_str());
             result.n_embd = -1;
             break;
         }
         if (ggml_n_dims(tensor) != 1) {
-            fprintf(stderr, "%s: invalid (non-1D) direction tensor shape in %s\n", __func__, load_info.fname.c_str());
+            LOG_ERR("%s: invalid (non-1D) direction tensor shape in %s\n", __func__, load_info.fname.c_str());
             result.n_embd = -1;
             break;
         }
@@ -2950,7 +1823,7 @@ static llama_control_vector_data llama_control_vector_load_one(const llama_contr
         if (result.n_embd == -1) {
             result.n_embd = ggml_nelements(tensor);
         } else if (ggml_nelements(tensor) != result.n_embd) {
-            fprintf(stderr, "%s: direction tensor in %s does not match previous dimensions\n", __func__, load_info.fname.c_str());
+            LOG_ERR("%s: direction tensor in %s does not match previous dimensions\n", __func__, load_info.fname.c_str());
             result.n_embd = -1;
             break;
         }
@@ -2967,7 +1840,7 @@ static llama_control_vector_data llama_control_vector_load_one(const llama_contr
     }
 
     if (result.n_embd == -1) {
-        fprintf(stderr, "%s: skipping %s due to invalid direction tensors\n", __func__, load_info.fname.c_str());
+        LOG_WRN("%s: skipping %s due to invalid direction tensors\n", __func__, load_info.fname.c_str());
         result.data.clear();
     }
 
@@ -2988,7 +1861,7 @@ llama_control_vector_data llama_control_vector_load(const std::vector<llama_cont
             break;
         }
         if (result.n_embd != -1 && result.n_embd != cur.n_embd) {
-            fprintf(stderr, "%s: control vectors in %s does not match previous dimensions\n", __func__, info.fname.c_str());
+            LOG_ERR("%s: control vectors in %s does not match previous dimensions\n", __func__, info.fname.c_str());
             result.n_embd = -1;
             break;
         }
@@ -3004,7 +1877,7 @@ llama_control_vector_data llama_control_vector_load(const std::vector<llama_cont
     }
 
     if (result.n_embd == -1) {
-        fprintf(stderr, "%s: no valid control vector files passed\n", __func__);
+        LOG_ERR("%s: no valid control vector files passed\n", __func__);
         result.data.clear();
     }
 
@@ -3075,7 +1948,7 @@ void yaml_dump_string_multiline(FILE * stream, const char * prop_name, const cha
 
 void yaml_dump_non_result_info(FILE * stream, const gpt_params & params, const llama_context * lctx,
                                const std::string & timestamp, const std::vector<int> & prompt_tokens, const char * model_desc) {
-    const llama_sampling_params & sparams = params.sparams;
+    const auto & sparams = params.sparams;
 
     fprintf(stream, "build_commit: %s\n",        LLAMA_COMMIT);
     fprintf(stream, "build_number: %d\n",        LLAMA_BUILD_NUMBER);
@@ -3095,6 +1968,7 @@ void yaml_dump_non_result_info(FILE * stream, const gpt_params & params, const l
     fprintf(stream, "cpu_has_sve: %s\n",         ggml_cpu_has_sve()         ? "true" : "false");
     fprintf(stream, "cpu_has_f16c: %s\n",        ggml_cpu_has_f16c()        ? "true" : "false");
     fprintf(stream, "cpu_has_fp16_va: %s\n",     ggml_cpu_has_fp16_va()     ? "true" : "false");
+    fprintf(stream, "cpu_has_riscv_v: %s\n",     ggml_cpu_has_riscv_v()     ? "true" : "false");
     fprintf(stream, "cpu_has_wasm_simd: %s\n",   ggml_cpu_has_wasm_simd()   ? "true" : "false");
     fprintf(stream, "cpu_has_blas: %s\n",        ggml_cpu_has_blas()        ? "true" : "false");
     fprintf(stream, "cpu_has_sse3: %s\n",        ggml_cpu_has_sse3()        ? "true" : "false");
@@ -3126,8 +2000,6 @@ void yaml_dump_non_result_info(FILE * stream, const gpt_params & params, const l
 
     fprintf(stream, "alias: %s # default: unknown\n", params.model_alias.c_str());
     fprintf(stream, "batch_size: %d # default: 512\n", params.n_batch);
-    yaml_dump_string_multiline(stream, "cfg_negative_prompt", sparams.cfg_negative_prompt.c_str());
-    fprintf(stream, "cfg_scale: %f # default: 1.0\n", sparams.cfg_scale);
     fprintf(stream, "chunks: %d # default: -1 (unlimited)\n", params.n_chunks);
     fprintf(stream, "color: %s # default: false\n", params.use_color ? "true" : "false");
     fprintf(stream, "ctx_size: %d # default: 512\n", params.n_ctx);
@@ -3138,10 +2010,7 @@ void yaml_dump_non_result_info(FILE * stream, const gpt_params & params, const l
     fprintf(stream, "grammar-file: # never logged, see grammar instead. Can still be specified for input.\n");
     fprintf(stream, "hellaswag: %s # default: false\n", params.hellaswag ? "true" : "false");
     fprintf(stream, "hellaswag_tasks: %zu # default: 400\n", params.hellaswag_tasks);
-
-    const auto logit_bias_eos = sparams.logit_bias.find(llama_token_eos(llama_get_model(lctx)));
-    const bool ignore_eos = logit_bias_eos != sparams.logit_bias.end() && logit_bias_eos->second == -INFINITY;
-    fprintf(stream, "ignore_eos: %s # default: false\n", ignore_eos ? "true" : "false");
+    fprintf(stream, "ignore_eos: %s # default: false\n", sparams.ignore_eos ? "true" : "false");
 
     yaml_dump_string_multiline(stream, "in_prefix", params.input_prefix.c_str());
     fprintf(stream, "in_prefix_bos: %s # default: false\n", params.input_prefix_bos ? "true" : "false");
@@ -3152,27 +2021,23 @@ void yaml_dump_non_result_info(FILE * stream, const gpt_params & params, const l
     fprintf(stream, "logdir: %s # default: unset (no logging)\n", params.logdir.c_str());
 
     fprintf(stream, "logit_bias:\n");
-    for (std::pair<llama_token, float> lb : sparams.logit_bias) {
-        if (ignore_eos && lb.first == logit_bias_eos->first) {
-            continue;
-        }
-        fprintf(stream, "  %d: %f", lb.first, lb.second);
+    for (const auto & logit_bias : sparams.logit_bias) {
+        fprintf(stream, "  %d: %f", logit_bias.token, logit_bias.bias);
     }
 
     fprintf(stream, "lora:\n");
-    for (std::tuple<std::string, float> la : params.lora_adapter) {
-        if (std::get<1>(la) != 1.0f) {
-            continue;
+    for (auto & la : params.lora_adapters) {
+        if (la.scale == 1.0f) {
+            fprintf(stream, "  - %s\n", la.path.c_str());
         }
-        fprintf(stream, "  - %s\n", std::get<0>(la).c_str());
     }
     fprintf(stream, "lora_scaled:\n");
-    for (std::tuple<std::string, float> la : params.lora_adapter) {
-        if (std::get<1>(la) == 1.0f) {
-            continue;
+    for (auto & la : params.lora_adapters) {
+        if (la.scale != 1.0f) {
+            fprintf(stream, "  - %s: %f\n", la.path.c_str(), la.scale);
         }
-        fprintf(stream, "  - %s: %f\n", std::get<0>(la).c_str(), std::get<1>(la));
     }
+    fprintf(stream, "lora_init_without_apply: %s # default: false\n", params.lora_init_without_apply ? "true" : "false");
     fprintf(stream, "main_gpu: %d # default: 0\n", params.main_gpu);
     fprintf(stream, "min_keep: %d # default: 0 (disabled)\n", sparams.min_keep);
     fprintf(stream, "mirostat: %d # default: 0 (disabled)\n", sparams.mirostat);
@@ -3210,7 +2075,6 @@ void yaml_dump_non_result_info(FILE * stream, const gpt_params & params, const l
 
     fprintf(stream, "rope_freq_base: %f # default: 10000.0\n", params.rope_freq_base);
     fprintf(stream, "rope_freq_scale: %f # default: 1.0\n", params.rope_freq_scale);
-    fprintf(stream, "seed: %u # default: -1 (random seed)\n", params.seed);
     fprintf(stream, "simple_io: %s # default: false\n", params.simple_io ? "true" : "false");
     fprintf(stream, "cont_batching: %s # default: false\n", params.cont_batching ? "true" : "false");
     fprintf(stream, "flash_attn: %s # default: false\n", params.flash_attn ? "true" : "false");
@@ -3220,11 +2084,11 @@ void yaml_dump_non_result_info(FILE * stream, const gpt_params & params, const l
     yaml_dump_vector_float(stream, "tensor_split", tensor_split_vector);
 
     fprintf(stream, "tfs: %f # default: 1.0\n", sparams.tfs_z);
-    fprintf(stream, "threads: %d # default: %u\n", params.n_threads, std::thread::hardware_concurrency());
+    fprintf(stream, "threads: %d # default: %u\n", params.cpuparams.n_threads, std::thread::hardware_concurrency());
     fprintf(stream, "top_k: %d # default: 40\n", sparams.top_k);
     fprintf(stream, "top_p: %f # default: 0.95\n", sparams.top_p);
     fprintf(stream, "min_p: %f # default: 0.0\n", sparams.min_p);
-    fprintf(stream, "typical_p: %f # default: 1.0\n", sparams.typical_p);
+    fprintf(stream, "typ_p: %f # default: 1.0\n", sparams.typ_p);
     fprintf(stream, "verbose_prompt: %s # default: false\n", params.verbose_prompt ? "true" : "false");
     fprintf(stream, "display_prompt: %s # default: true\n", params.display_prompt ? "true" : "false");
 }