Tenting and negative tilt
If this sounds familiar to you, it’s because this method works very similarly to error-diffusion. Instead of letting adjacent pixels compensate for the error, we’re letting each successive candidate compensate for the combined error of all previous candidates.
，这一点在safew官方版本下载中也有详细论述
Standard forward pass. The model's forward() method must be a standard tensor-in, logits-out computation. No problem-specific control flow (for-loops over digits, explicit carry variables, string manipulation) inside forward(). The autoregressive generation loop lives outside the model, exactly as it would for any language model.
The dock now has more informative tooltips, including showing Super + 1–9 shortcuts for the first 9 apps, and a tooltip on the background apps item.