Ablations with InfoNCE
#5
by
tomaarsen
HF staff
- opened
Hello!
I'm quite interested in your work - I think progressively shifting the focus of the learning is quite smart. I was wondering if you performed any ablations against just InfoNCE? It's a bit hard to know whether your loss function actually outperforms InfoNCE, or if the gains are e.g. just from using better datasets.
- Tom Aarsen
Same concern