Method build_inp_attn_scale

subprojects/llama.cpp/src/llama-graph.cpp:1460–1472

}

ggml_tensor * llm_graph_context::build_inp_attn_scale() const {
    auto inp = std::make_unique<llm_graph_input_attn_temp>(hparams.n_attn_temp_floor_scale, hparams.f_attn_temp_scale, hparams.f_attn_temp_offset);

    auto & cur = inp->attn_scale;

    // this need to be 1x1xN for broadcasting
    cur = ggml_new_tensor_3d(ctx0, GGML_TYPE_F32, 1, 1, n_tokens);
    ggml_set_input(cur);

    res->add_input(std::move(inp));

    return cur;
}

ggml_tensor * llm_graph_context::build_inp_out_ids() const {
    // note: when all tokens are output, we could skip this optimization to spare the ggml_get_rows() calls,