Method build_lora_mm_id

subprojects/llama.cpp/src/llama-graph.cpp:851–877

}

ggml_tensor * llm_graph_context::build_lora_mm_id(
          ggml_tensor * w,   // ggml_tensor * as
          ggml_tensor * cur, // ggml_tensor * b
          ggml_tensor * ids) const {
    ggml_tensor * res = ggml_mul_mat_id(ctx0, w, cur, ids);
    for (const auto & lora : *loras) {
        llama_adapter_lora_weight * lw = lora.first->get_weight(w);
        if (lw == nullptr) {
            continue;
        }

        const float alpha = lora.first->alpha;
        const float rank  = (float) lw->b->ne[0];
        const float scale = alpha ? lora.second * alpha / rank : lora.second;

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

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

    return res;
}

ggml_tensor * llm_graph_context::build_norm(
         ggml_tensor * cur,