def call(self, q_reps, p_reps):
if self.negatives_cross_device:
# This gathers both negatives and positives.
# It could likely be optimized by only gathering negatives.
q_reps = self._dist_gather_tensor(q_reps)
p_reps = self._dist_gather_tensor(p_reps)
scores = self.compute_similarity(q_reps, p_reps) / self.temperature
scores = scores.view(q_reps.size(0), -1)
target = torch.arange(scores.size(0), device=scores.device, dtype=torch.long)
target *= (p_reps.size(0) // q_reps.size(0))
return self.cross_entropy(scores, target)
in the code，does it use ContrastiveLoss following the paper？
def call(self, q_reps, p_reps):
if self.negatives_cross_device:
# This gathers both negatives and positives.
# It could likely be optimized by only gathering negatives.
q_reps = self._dist_gather_tensor(q_reps)
p_reps = self._dist_gather_tensor(p_reps)
scores = self.compute_similarity(q_reps, p_reps) / self.temperature
scores = scores.view(q_reps.size(0), -1)
in the code，does it use ContrastiveLoss following the paper？