tinydistill live
connecting...
waiting
no status yet
Speed over this stage
The teacher's labels so far
Training the student
starts after labelling
Loss (distillation loss: lower is closer to the teacher)
Agreement with the teacher's top label
Learning rate
Solid: on the training batches, every few batches. Dashed with dots: on the
held-out texts, after each epoch (with --val-share). Epochs on the x axis.
Per epoch, as a table
Stages done
none yet
Results
written when the run is evaluated
The full results (results.md):
every model, per-label scores, speed and the cascade.