jbm

Thanks so much for the reply! Digging around the solver code (as you pointed out) it did seem like the joint embedding might want a prompt, so I did add...

Actually though... what determines _when_ it will generate a sample output? I can see it running through train and valid steps, and it's saving checkpoints, but I don't seem to...

Yes, I saw from another Issue/comment that the "every" in the "generate" config refers to epochs, not steps. I had it set to 1000, thinking it meant step, so I...

I'm having the same issue—did you manage to resolve this? I have also tried adding the `-d` flag, btw, but same error.

I installed using the 2.3.x version, cuda 12.1, conda (from the DGL website), with torch 2.3.0, and I'm seeing: ``` >>> import dgl Traceback (most recent call last): File "",...

Yeah, this is a terrible policy (and customer experience) for sure. I think the biggest issue is that it _seems_ like compacting history gets included. I suppose it makes sense,...

It's just frustrating, that's all. I think the thing that bothers me is that we're paying for chat context length and for compacting when contexts get too long. I've mitigated...

Look, I completely get your perspective. And you're right; in reality, it is already far too cheap, according to the cost of providing the product. This is a typical race-to-the-bottom,...

"What do you actually mean? What is it that you think are paying for? And what do people expect to get from it?" To be blunt, premature over-investment in Transformers...