SlotFlow Pretrained Model v1.0.0

This release provides the official pretrained SlotFlow model for sinusoidal posterior inference and variable-cardinality estimation.
The model was trained on a large-scale 8-million–sample MultiSinusoidDataset, spanning amplitudes, frequencies, phases, component counts, frequency separations, and noise realizations as described the official publication.

Using this pretrained checkpoint, users can immediately:

• run test-time inference without retraining,
• reproduce the key qualitative behaviors shown in the repository (e.g., posterior samples, slot consistency, reconstructions),
  and verify quantitative results such as cardinality calibration, RMSE curves, and frequency-separation performance.

In short, this release enables full out-of-the-box evaluation of SlotFlow exactly as demonstrated in the provided notebook Eval.py.