How to convert KataGo .bin network file to PyTorch .pth?

[ZGJ605]

No description provided.

[isty2e]

convert.py already can load the (v8) .bin model into PyTorch. You can just torch.save() with appropriate arguments to save the model as .pt or .pth.

[ZGJ605]

Does it support b6c96?

[ZGJ605]

I can not load https://d3dndmfyhecmj0.cloudfront.net/g170/neuralnets/index.html. successfully.And what is config file?

[isty2e]

You can download some networks from https://katagotraining.org/networks/ (note that only v8 networks are supported for the time being so most recent ones are not supported). Once you unzip the .gz file, you'll see there is a .json config file.

[ZGJ605]

I have found networks .But I can not find the config file.

[isty2e]

Extract the .bin.gz file first.

[ZGJ605]

I have extracted but not found.

[isty2e]

Then download the corresponding .zip file (Raw TF weights) and extract it. model.config.json is in there.

[ZGJ605]

How can I change the board size?