Janek Zimoch
Janek Zimoch
My work around is to: 1. save topic embeddings during training 2. during inference, after adding new documents, do topic assignment myself by finding closest topic embedding to a given...
I would be also interested in getting to know what might be causing this issue. I had similar problem. I tried to recognise between different types of Gatorade isotonic drinks...
I tried recreating this as well, and I think the output is deterministic, so I don't think there is any bug on the DB side. I think what @moladeyal saw...