|
|
|
@ -760,9 +760,9 @@ class TrainingState():
|
|
|
|
|
|
|
|
|
|
self.metrics['step'] = [f"{self.epoch}/{self.epochs}"]
|
|
|
|
|
if self.epochs != self.its:
|
|
|
|
|
self.metric.append(f"{self.it}/{self.its}")
|
|
|
|
|
self.metrics.append(f"{self.it}/{self.its}")
|
|
|
|
|
if steps > 1:
|
|
|
|
|
self.metric.append(f"{step}/{steps}")
|
|
|
|
|
self.metrics.append(f"{step}/{steps}")
|
|
|
|
|
self.metrics['step'] = ", ".join(self.metrics['step'])
|
|
|
|
|
|
|
|
|
|
if lapsed:
|
|
|
|
@ -818,7 +818,7 @@ class TrainingState():
|
|
|
|
|
deriv = 0
|
|
|
|
|
accum_length = len(self.losses)//2 # i *guess* this is fine when you think about it
|
|
|
|
|
loss_value = self.losses[-1]["value"]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
for i in range(accum_length):
|
|
|
|
|
d1_loss = self.losses[accum_length-i-1]["value"]
|
|
|
|
|
d2_loss = self.losses[accum_length-i-2]["value"]
|
|
|
|
|