Skip to content

Lean-Contextualize builds dependency-aware Lean contexts from Jixia symbol/decl graphs. It resolves constructors, instances, and wrappers to produce the minimal context required for compiling individual problems. This is designed to enhance dataset generation from custom Lean4 codebases.

License

Notifications You must be signed in to change notification settings

ataylor24/Lean-Contextualize

Repository files navigation

Lean-Contextualize

Lean-Contextualize builds dependency-aware Lean contexts from Jixia symbol/decl graphs. It resolves constructors, instances, and wrappers to produce the minimal context required for compiling individual problems. This codebase designed to enhance dataset generation from custom Lean4 codebases.

Usage:

  1. Make sure Jixia and custom Lean project (ex. analysis) lean versions match (current: version 4.23.0-rc2).
  2. Run orchestrate_jixia.py to construct the processed analysis files.
  3. Run build_context.py to perform type and value resolution and generate the initial version of the generated context JSONL.
  4. Verify each row.
  5. Run second_pass_api_check.py to query the Open-AI API to correct errors (default: GPT-5-Medium).

About

Lean-Contextualize builds dependency-aware Lean contexts from Jixia symbol/decl graphs. It resolves constructors, instances, and wrappers to produce the minimal context required for compiling individual problems. This is designed to enhance dataset generation from custom Lean4 codebases.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published