Method load_models

subprojects/llama.cpp/tools/server/server-models.cpp:197–308

TODO: allow refreshing cached model list

// TODO: allow refreshing cached model list
void server_models::load_models() {
    // loading models from 3 sources:
    // 1. cached models
    common_presets cached_models = ctx_preset.load_from_cache();
    SRV_INF("Loaded %zu cached model presets\n", cached_models.size());
    // 2. local models from --models-dir
    common_presets local_models;
    if (!base_params.models_dir.empty()) {
        local_models = ctx_preset.load_from_models_dir(base_params.models_dir);
        SRV_INF("Loaded %zu local model presets from %s\n", local_models.size(), base_params.models_dir.c_str());
    }
    // 3. custom-path models from presets
    common_preset global = {};
    common_presets custom_presets = {};
    if (!base_params.models_preset.empty()) {
        custom_presets = ctx_preset.load_from_ini(base_params.models_preset, global);
        SRV_INF("Loaded %zu custom model presets from %s\n", custom_presets.size(), base_params.models_preset.c_str());
    }

    // cascade, apply global preset first
    cached_models  = ctx_preset.cascade(global, cached_models);
    local_models   = ctx_preset.cascade(global, local_models);
    custom_presets = ctx_preset.cascade(global, custom_presets);

    // note: if a model exists in both cached and local, local takes precedence
    common_presets final_presets;
    for (const auto & [name, preset] : cached_models) {
        final_presets[name] = preset;
    }
    for (const auto & [name, preset] : local_models) {
        final_presets[name] = preset;
    }

    // process custom presets from INI
    for (const auto & [name, custom] : custom_presets) {
        if (final_presets.find(name) != final_presets.end()) {
            // apply custom config if exists
            common_preset & target = final_presets[name];
            target.merge(custom);
        } else {
            // otherwise add directly
            final_presets[name] = custom;
        }
    }

    // server base preset from CLI args take highest precedence
    for (auto & [name, preset] : final_presets) {
        preset.merge(base_preset);
    }

    // convert presets to server_model_meta and add to mapping
    for (const auto & preset : final_presets) {
        server_model_meta meta{
            /* preset       */ preset.second,
            /* name         */ preset.first,
            /* port         */ 0,
            /* status       */ SERVER_MODEL_STATUS_UNLOADED,
            /* last_used    */ 0,