medium_final — 3 epochs done

trainer_state.json

@@ -270,6 +270,15 @@
       "learning_rate": 1.474056603773585e-06,
       "loss": 20.45797607421875,
       "step": 1700
+    },
+    {
+      "epoch": 3.0,
+      "step": 1749,
+      "total_flos": 1.48405163655168e+16,
+      "train_loss": 20.713123458940824,
+      "train_runtime": 4479.5407,
+      "train_samples_per_second": 24.963,
+      "train_steps_per_second": 0.39
     }
   ],
   "logging_steps": 50,