looskis/tarski
Reading diffusion language models for decisions: same OpenJev model, one pass instead of 3.65, +5 points, calibrated, no training. Plus cheap task branches on a frozen encoder.
GitHub repository with 25 stars and 0 forks.
Language: Python