Pipeline era from directions
We implement InstructPipe with a two-stage LLM refinement prompting technique, adopted by a pseudocode interpretation step to render a pipeline. The determine under illustrates the high-level workflow of the InstructPipe implementation. InstructPipe leverages two LLM modules (highlighted in pink) — a Node Selector, and a Code Author. Given a consumer instruction and a pipeline tag (e.g., a multimodal pipeline), we first devise the Node Selector to determine a listing of potential nodes that may be used based on the directions. Within the Node Selector, we immediate the LLM with a really temporary description of every node, aiming to filter out unrelated nodes for a goal pipeline. The chosen nodes and the unique consumer enter (the immediate and the tag) are then fed into the Code Author, which generates pseudocode (i.e., a succinct code format that defines the choices and connections of the important nodes) for the specified pipeline. In Code Author, we offer the LLM with detailed descriptions and examples of every chosen node to make sure the LLM has in depth context for every candidate node. Lastly, we make use of a Code Interpreter to parse the pseudocode and render a visible programming pipeline with which the consumer could work together.