Beta 3 model

This is the same model used in my latest dev playthrough!

These models can be used to power the EndlessAWSW server. They are heavily optimized to work on CPU.

At the time of writing, I decided to focus on CPU support due to the chip shortage in the world right now - it's just cheaper for me to host on a server with CPU I already have than to get GPU's for it, and it was reasonably doable.

Unlike Beta 1, there's only 1 model:

• Normal - This is the "as-is" model, the only change is that it's converted from PyTorch from ONNX for accelerated CPU support, but no loss in quality.

Pretrained Models

These models can be used to power the EndlessAWSW server. They are heavily optimized to work on CPU.

At the time of writing, I decided to focus on CPU support due to the chip shortage in the world right now - it's just cheaper for me to host on a server with CPU I already have than to get GPU's for it, and it was reasonably doable.

There are 2 different models:

• Tiny - This model is fast but less accurate. While it still delivers on a good level of quality, you can expect better results with the normal model. The tiny model is also used in the public servers (at least the ones I host). The tiny model is about 10 times faster due to the use of int8's rather than float32's. On CPU, it brings a lot of performance at a slight trade off for accuracy.
• Normal - This is the "as-is" model, the only change is that it's converted from PyTorch from ONNX for accelerated CPU support, but no loss in quality. It is considerably slower, but it's something you might want to try if you wish to run it on your own.