Designer Design Document

Overview

Designer is a core component in the AI Kernel Generator that automatically generates algorithm design documents based on Large Language Models (LLMs). It inherits from AgentBase and is responsible for intelligently generating high-quality algorithm sketches based on the kernel name, task description, and hardware configuration. The Designer uses AUL (AI Unity Language) or similar design languages to express algorithm logic.

Core Functions

• Intelligent Design Generation: Automatically generates algorithm design documents based on kernel requirements
• Doc-Driven Integration: Supports custom reference documents to improve generation quality
• Multi-DSL Support: Supports different design languages
• Hardware-Aware Design: Considers hardware characteristics during design generation
• Document Integration: Automatically loads design specifications and reference materials

Initialization Parameters

Parameter Name	Type/Required	Description
op_name	str (Required)	Kernel name, identifying the specific kernel
task_desc	str (Required)	Task description, detailing the kernel functional requirements
dsl	str (Required)	Design language: "triton_cuda", "triton_ascend", "swft", etc.
backend	str (Required)	Hardware backend: "ascend", "cuda", etc.
arch	str (Required)	Hardware architecture: "ascend910b4", "a100", etc.
workflow_config_path	str (Optional)	Workflow configuration file path (usually injected by Task from the orchestration plan)
config	dict (Required)	Complete orchestration plan configuration including log_dir, agent_model_config, docs_dir, etc. (see Task Orchestration Plan Configuration)

Related docs: Workflow details in Workflow; documentation integration in Doc-Driven Integration Guide.

Doc-Driven Integration

The Designer loads reference documents via docs_dir.designer in the orchestration plan. For document list and specifications, see the Doc-Driven Integration Guide. Details are not duplicated here.

Execution Flow

1. Initialization Stage

   • Load workflow configuration and create parser
   • Initialize design generation template
   • Load reference documents using Doc-Driven Integration
   • Prepare base document structure

2. Generation Stage

   • Process task information and conductor suggestions
   • Execute LLM generation using loaded documents
   • Return generated design, prompt, and reasoning

3. Document Structure

   • DSL specifications and syntax rules
   • Algorithm design patterns and examples
   • Hardware-specific considerations
   • Format instructions for output parsing