Recipe Catalog Generation

Relevant source files

The following files were used as context for generating this wiki page:

• src/data/olive-recipes-catalog.ts
• src/components/features/InputEnvironmentPanel.tsx
• src/lib/oliveRecipeHub.ts (Inferred from component imports)
• package.json
• README.md
• CONTRIBUTING.md
• scripts/generate-olive-recipes-catalog.mjs (Referenced in automation scripts)

Recipe Catalog Generation is the automated process of synchronizing, parsing, and surfacing optimized model configuration presets from the official microsoft/olive-recipes repository into Olive Studio. This system enables users to discover and apply verified optimization pipelines for specific model architectures and hardware targets without manual JSON configuration.

Sources: README.md:144-148, CONTRIBUTING.md:73-77

Architecture and Data Flow

The catalog generation system follows a periodic update cycle where upstream GitHub data is transformed into a local TypeScript registry. At runtime, this registry is used to lazy-load the actual recipe payloads directly from GitHub.

Catalog Generation Pipeline

The generation process is triggered by a maintenance script that scrapes the upstream repository structure. This script creates a metadata index used by the UI to filter and search for recipes.

The system differentiates between Metadata Source types:

• recipe: Metadata is derived directly from the recipe content.
• folder: Metadata (architecture and device) is inferred from the repository directory structure.

Sources: src/data/olive-recipes-catalog.ts:7-14, src/components/features/InputEnvironmentPanel.tsx:180-184, CONTRIBUTING.md:77

Data Structures and Configuration

The catalog is stored as a constant array of RecipeCatalogItem objects, which serve as the source of truth for the "Starter Curated" presets in the UI.

RecipeCatalogItem Schema

Field	Type	Description
name	string	Display name, typically formatted as model · device · filename.
architecture	string	Model family (e.g., Llama, Phi, Whisper, BERT).
device	string	Target hardware/EP (e.g., CPU, CUDA, QNN, OpenVINO).
repoPath	string	Relative path to the JSON file in the upstream repo.
description	string	Human-readable summary of the optimization goal.
metadataSource	"folder" | "recipe"	Origin of the architecture/device tags.

Sources: src/data/olive-recipes-catalog.ts:14-23, src/lib/oliveRecipeHub.ts

Global Repository Constants

The system is hardcoded to track a specific upstream target to ensure compatibility:

• Repository: microsoft/olive-recipes
• Branch: main

Sources: src/data/olive-recipes-catalog.ts:6-7

Maintenance and Automation

The catalog is not intended to be updated manually. Developers use built-in scripts to synchronize the local metadata with the latest upstream additions.

Update Workflow

1. Run npm run generate:recipes to execute the generation script.
2. The script updates src/data/olive-recipes-catalog.ts.
3. The UI automatically reflects new models upon the next build or hot reload.

Integration with Runtime Hardware Probing

When a catalog item is selected, the system compares the recipe's requirements against the local machine's capabilities.

Compatibility Matching Logic

The InputEnvironmentPanel uses the catalog metadata to:

1. Filter: Hide recipes targeting NPUs (like QNN) or specific GPUs if the hardware probe does not detect them.
2. Warn: Surface "Incompatible" badges if a user selects a recipe for DirectML on a Linux machine.
3. Hardware-Aware Display: Highlight "Compatible" presets based on detected Execution Providers (EPs).

Sources: src/components/features/InputEnvironmentPanel.tsx:340-355, src/components/features/InputEnvironmentPanel.tsx:505-520

Example Catalog Entry

{
  name: "deepseek-ai-DeepSeek-R1-Distill-Qwen-1.5B · QNN · config_gpu",
  architecture: "Qwen",
  device: "QNN",
  repoPath: "deepseek-ai-DeepSeek-R1-Distill-Qwen-1.5B/QNN/config_gpu.json",
  description: "Official microsoft/olive-recipes run config for deepseek-ai-DeepSeek-R1-Distill-Qwen-1.5B (QNN/config_gpu.json).",
  metadataSource: "folder",
}

Sources: src/data/olive-recipes-catalog.ts:167-174

Summary

The Recipe Catalog Generation system provides a bridge between the rapidly evolving microsoft/olive-recipes ecosystem and the Olive Studio UI. By automating the extraction of metadata and providing a structured TypeScript registry, the project ensures that users have access to hundreds of verified optimization pipelines for modern LLMs and vision models while maintaining strict hardware compatibility checks.