Skip to content

v1.0.0

Choose a tag to compare

View all tags
@dev-kas dev-kas released this 26 Jan 19:09
· 4 commits to master since this release
v1.0.0
c4190c5

Release v1.0.0

Key Highlights

  • Scheduler engine: Uses a Temporal Convolutional Network (TCN) to dual-predict both the next key and the specific timing (delta/duration) of the press.
  • Realistic error injection: Simulates "fat-finger" typos based on actual QWERTY keyboard adjacency: the model doesn't just make mistakes, it makes human mistakes.
  • Full data pipeline: Includes a curses-based TUI recorder for capturing your own typing style and a synthesizer to generate massive amounts of training data.

Licensing (Hybrid)

This project uses a dual-license structure:

  • Code: Licensed under AGPLv3.
  • Model Artifacts: The pre-trained weights (base.pth) and tokenizer (tokens.json) are under CC-BY-NC 4.0.

Quick start

To see the model in action immediately, run:

python -m examples.typewriter

What's included:

  • base.pth & tokens.json: The pre-trained brain of the project
  • model/: TCN architecture and training logic
  • recorder/: Everything needed to capture your own typing styles
  • examples/: Real-world usage scripts
Assets 4
Loading