Method build_attn_inp_cross

subprojects/llama.cpp/src/llama-graph.cpp:2043–2054

}

llm_graph_input_attn_cross * llm_graph_context::build_attn_inp_cross() const {
    auto inp = std::make_unique<llm_graph_input_attn_cross>(cross);

    const int32_t n_enc = !cross->v_embd.empty() ? cross->n_enc : hparams.n_ctx_train;

    inp->cross_kq_mask = ggml_new_tensor_4d(ctx0, GGML_TYPE_F32, n_enc, n_tokens, 1, 1);
    ggml_set_input(inp->cross_kq_mask);

    inp->cross_kq_mask_cnv = cparams.flash_attn ? ggml_cast(ctx0, inp->cross_kq_mask, GGML_TYPE_F16) : inp->cross_kq_mask;

    return (llm_graph_input_attn_cross *) res->add_input(std::move(inp));
}

ggml_tensor * llm_graph_context::build_attn(
        llm_graph_input_attn_cross * inp,