Function make_decoder_state

src/models/pocket_tts/mimi_decoder.cpp:1087–1114

    int64_t context_ = 0;
    int64_t cache_steps_ = 0;
    int64_t head_dim_ = 0;
    ggml_context * ggml_ctx_ = nullptr;
    ggml_backend_buffer_t params_buffer_ = nullptr;
    ggml_cgraph * graph_ = nullptr;
    ggml_gallocr_t galloc_ = nullptr;
    core::TensorValue input_bct_;
    core::TensorValue positions_;
    core::TensorValue attention_mask_;
    std::vector<core::TensorValue> attention_masks_;
    std::vector<std::vector<float>> attention_mask_values_;
    std::vector<core::TensorValue> prefix_keys_;
    std::vector<core::TensorValue> prefix_values_;
    std::vector<core::TensorValue> work_prefix_keys_;
    std::vector<core::TensorValue> work_prefix_values_;
    std::vector<core::TensorValue> zero_prefix_keys_;
    std::vector<core::TensorValue> zero_prefix_values_;
    std::vector<core::TensorValue> keys_;
    std::vector<core::TensorValue> values_;
    std::vector<int32_t> positions_buffer_;
    std::vector<std::vector<ggml_tensor *>> carry_key_sources_;
    std::vector<std::vector<ggml_tensor *>> carry_value_sources_;
    std::vector<std::vector<ggml_tensor *>> carry_key_destinations_;
    std::vector<std::vector<ggml_tensor *>> carry_value_destinations_;
    std::vector<std::vector<ggml_tensor *>> append_key_sources_;
    std::vector<std::vector<ggml_tensor *>> append_value_sources_;
    std::vector<std::vector<ggml_tensor *>> append_key_destinations_;
    std::vector<std::vector<ggml_tensor *>> append_value_destinations_;
    core::TensorValue output_bct_;
    int64_t current_end_ = 0;
    int64_t runtime_keep_steps_ = 0;
    int64_t current_mask_keep_steps_ = -1;