name: "screen_sbert" platform: "ensemble" input [ { name: "IMAGE_BYTES" data_type: TYPE_STRING dims: [1] } ] output [ { name: "screen_embedding" data_type: TYPE_FP32 dims: [-1, 512] } ] ensemble_scheduling { step { model_name: "parse_gui" model_version: -1 input_map { key: "IMAGE_BYTES" value: "IMAGE_BYTES" } output_map { key: "bbox" value: "bbox_2d" } output_map { key: "function_embedding" value: "function_embedding_2d" } output_map { key: "vision_embedding" value: "vision_embedding_2d" } output_map { key: "text_embedding" value: "text_embedding_2d" } } step { model_name: "screen_sbert_adapter" model_version: -1 input_map { key: "bbox_2d" value: "bbox_2d" } input_map { key: "function_embedding_2d" value: "function_embedding_2d" } input_map { key: "vision_embedding_2d" value: "vision_embedding_2d" } input_map { key: "text_embedding_2d" value: "text_embedding_2d" } output_map { key: "bbox" value: "bbox_3d" } output_map { key: "function_embedding" value: "function_embedding_3d" } output_map { key: "vision_embedding" value: "vision_embedding_3d" } output_map { key: "text_embedding" value: "text_embedding_3d" } output_map { key: "padding_mask" value: "padding_mask_2d" } } step { model_name: "screen_sbert_engine" model_version: -1 input_map { key: "bbox" value: "bbox_3d" } input_map { key: "function_embedding" value: "function_embedding_3d" } input_map { key: "vision_embedding" value: "vision_embedding_3d" } input_map { key: "text_embedding" value: "text_embedding_3d" } input_map { key: "padding_mask" value: "padding_mask_2d" } output_map { key: "screen_embedding" value: "screen_embedding" } } }