A high-performance inference engine for AI models
#include "util.hpp"
#include <arg.h>
#include <common.h>
#include <ggml-cpp.h>
#include <stdexcept>
#include <string_view>
#include <vector>
std::filesystem::path get_model_path(
const std::string& model,
bool dflash
) {
if (std::filesystem::is_regular_file(model)) {
return model;
}
common_params params;
auto& model_params = dflash ? params.speculative.draft.mparams : params.model;
model_params.hf_repo = model;
if (dflash) {
// DFlash sidecars are excluded from the upstream primary-model search.
params.speculative.types = {COMMON_SPECULATIVE_TYPE_DRAFT_DFLASH};
}
auto handler = common_models_handler_init(params, LLAMA_EXAMPLE_COMMON);
common_models_handler_apply(handler, params);
if (model_params.path.empty()) {
throw std::runtime_error("No model found in " + model);
}
return model_params.path;
}