DaehanKim

What is the purpose of reconstructing your adjacency matrix? Since you have a sparse graph, there are not much training signals for some nodes, resulting in inaccurate edge reconstruction. Maybe...

Why don't you use true adjacency matrix as a reward signal, instead of reconstructing it? I don't have much to tell about auxiliary approaches since I have no clue on...

I also observed a spike in policy loss when running sentiment-control example, and I initially thought it's because of some strange samples or high variance in positive logits. And I...

I corrected parsing pipeline output and loss spike still remains in sentiment-control notebook example. so there may be another reaseon for this unstability. 

1. By 'negative sampling', do you mean balancing each class (with same number of examples) in multi-class setting? 2. With a small number of training epochs, does the model converge?...

Thanks for the reply. I thought that the **positive weight** in your implementation was to balance positive edges against negative edges. As I see it, this would serve as a...

Thanks for all detailed replies. I'm looking forward to the release of your reference code for node classification.

imo, residual clipping seems beneficial to prevent policy loss spiking reported in #101 . It's probably coming from instability in value estimation.

안녕하세요! koalpaca를 공개해 주셔서 감사합니다. 저도 polyglot-5.8b를 백본으로 사용한 koalpaca 모델에 inference를 해보고 있는데 특별히 사용하신 generation parameter가 있으신지 궁금합니다. 아래 코드로도 생성은 되는데 polyglot 예시에 대해서는 polyglutino로 단어가 바뀌는...