Skip to content

Releases: troyscott/markforge

Release list

MarkForge v0.2.1

MarkForge v0.2.1 Pre-release
Pre-release

Choose a tag to compare

View all tags
@troyscott troyscott released this 24 Aug 03:21
v0.2.1
a726108

MarkForge v0.2.1

This prerelease packages MarkForge for direct installation while preserving its existing commands and Python import name.

Packaging

  • Distribution name: markforge-docs
  • Python package: markforge
  • Console commands: markforge and markforge-gui
  • Python requirement: 3.12
  • Includes a universal Python wheel and source archive
  • Adds public project metadata and package building to the cross-platform CI matrix

Installation from this release

uv tool install https://github.com/troyscott/markforge/releases/download/v0.2.1/markforge_docs-0.2.1-py3-none-any.whl
markforge --help

Marker may download local model files during the first PDF conversion. On macOS and CPU-only Linux, install llama.cpp for difficult OCR and layout recovery.

Validation

  • 12 tests pass locally
  • The wheel installs and launches in an isolated environment
  • Package tests and builds pass on GitHub Actions for Linux, macOS, and Windows
  • Real Marker conversion remains verified on Apple MPS and CPU

Windows packaging is CI-verified, but NVIDIA CUDA inference still requires a real hardware smoke test.

Loading

MarkForge v0.2.0

MarkForge v0.2.0 Pre-release
Pre-release

Choose a tag to compare

View all tags
@troyscott troyscott released this 24 Aug 03:11
v0.2.0
a121ddc

MarkForge v0.2.0

This prerelease rebuilds MarkForge as a cross-platform, chapter-aware document conversion utility.

Highlights

  • Shared conversion engine used by the CLI and CustomTkinter desktop application
  • Chapter-level PDF output using bookmarks, detected headings, or fixed page ranges
  • Smaller internal processing chunks with adaptive splitting and CPU fallback
  • Current Marker PdfConverter API for PDFs
  • MarkItDown support for DOCX, PPTX, and XLSX
  • Apple MPS, NVIDIA CUDA, and CPU device selection
  • Per-document manifests, atomic writes, resume, forced reprocessing, and safe temporary cleanup
  • Collision-safe image extraction with relative Markdown links
  • Optional combined Markdown books
  • Reproducible Python 3.12 project workflow using uv and a committed lockfile
  • GitHub Actions test matrix for macOS, Windows, and Linux

Validation

  • Unit and integration suite passes with 12 tests
  • Marker conversion verified locally on Apple MPS and CPU
  • A representative 42-page Microsoft Fabric chapter was converted with Marker, including tables and 31 linked images
  • Full 754-page chapter coverage was validated with the embedded-text extractor

Prerelease note

Windows and NVIDIA CUDA compatibility is preserved in the implementation and CI, but a real Windows/NVIDIA hardware smoke test has not yet been completed.

Source documents, converted books, model caches, and other private artifacts are not included in this release.

Loading