I’m trying to finetune Llama 2 model using DPO on a multi GPU set up. The setup has two V100s. The code’s structure is following while omitting unnecessary code # Imports device = {"": ...
In: from geomloss import SamplesLoss In: loss = SamplesLoss(loss="sinkhorn", p=2, blur=.05) In: a = b = Tensor([[1, 2]]) In: loss(a, b) Out: sinkhorn_divergence.py ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results
Feedback