Method build_lora_mm

subprojects/llama.cpp/src/llama-graph.cpp:825–849

}

ggml_tensor * llm_graph_context::build_lora_mm(
          ggml_tensor * w,
          ggml_tensor * cur) const {
    ggml_tensor * res = ggml_mul_mat(ctx0, w, cur);

    for (const auto & lora : *loras) {
        llama_adapter_lora_weight * lw = lora.first->get_weight(w);
        if (lw == nullptr) {
            continue;
        }

        const float adapter_scale = lora.second;
        const float scale = lw->get_scale(lora.first->alpha, adapter_scale);

        ggml_tensor * ab_cur = ggml_mul_mat(
                ctx0, lw->b,
                ggml_mul_mat(ctx0, lw->a, cur)
                );

        ab_cur = ggml_scale(ctx0, ab_cur, scale);
        res = ggml_add(ctx0, res, ab_cur);
    }

    return res;
}

ggml_tensor * llm_graph_context::build_lora_mm_id(
          ggml_tensor * w,   // ggml_tensor * as