diff --git a/Playgrounds/prova.py b/Playgrounds/prova.py
index f05f239..ef42ac9 100644
--- a/Playgrounds/prova.py
+++ b/Playgrounds/prova.py
@@ -1,4 +1,5 @@
 import random
+import time
 import torch
 import pandas as pd
 from pathlib import Path
@@ -26,7 +27,7 @@ TOKENANO = BPE.TokeNanoCore(VOCABULARY, SPECIAL_VOC)
 # Constants
 TOKEN_SPACE_SIZE = TOKENANO.vocabulary_size + 1
 EMBEDDED_SIZE = 256
-FEED_FORWARD_MULTIPLIER =  4
+FEED_FORWARD_MULTIPLIER = 4
 ATTENTION_HEADS = 8
 SENTENCE_LENGTH = 256
 NUMBER_OF_BLOCKS = 4
@@ -45,7 +46,7 @@ TOY_DATASET = pd.read_csv(TOY_DATASET_PATH)
 TOY_BATCH_INPUT_LIST: list[list[int]] = []
 TOY_BATCH_PADDING_LIST: list[list[bool]] = []
 TOY_BATCH_TARGET_LIST: list[list[int]] = []
-TOY_BATCH_DECODER_DEFAULT: list[list[int]]= []
+TOY_BATCH_DECODER_DEFAULT: list[list[int]] = []
 
 
 for index, row in TOY_DATASET.iterrows():
@@ -98,7 +99,7 @@ NANOSOCRATES = Transformer.TrainingModel(
     EMBEDDED_SIZE,
     FEED_FORWARD_MULTIPLIER,
     ATTENTION_HEADS,
-    NUMBER_OF_BLOCKS
+    NUMBER_OF_BLOCKS,
 )
 cross_entropy = torch.nn.CrossEntropyLoss(ignore_index=PAD_TOKEN)
 optimizer = torch.optim.AdamW(NANOSOCRATES.parameters())
@@ -120,21 +121,31 @@ while current_epoch < MAX_EPOCHS:
     last_loss = 0
     last_prediction: torch.Tensor
 
+    LOSS_HISTORY = []
+
+    start = time.time_ns()
+
+
     for i in range(0, SENTENCE_LENGTH):
 
+
         optimizer.zero_grad()
         tgt_padding = decoder_list.eq(PAD_TOKEN)
 
-        logits: torch.Tensor = NANOSOCRATES((encoder_list, src_padding, decoder_list, tgt_padding))
+        logits: torch.Tensor = NANOSOCRATES(
+            (encoder_list, src_padding, decoder_list, tgt_padding)
+        )
         prob = torch.softmax(logits, 2)
 
         most_probable_tokens = torch.argmax(prob, 2)
         last_prediction = most_probable_tokens
 
-        logits = logits[:,:i,:]
-        logits = logits.permute(0, 2, 1)
+        logits = logits[:, i, :]
+        # logits = logits.permute(0, 2, 1)
 
-        loss : torch.Tensor = cross_entropy(logits, target_logits[:, 0:i])
+        loss: torch.Tensor = cross_entropy(logits, target_logits[:, i])
+        LOSS_HISTORY.append(loss.item())
+        # loss : torch.Tensor = cross_entropy(logits, target_logits[:, 0:i])
         # loss : torch.Tensor = cross_entropy(logits, target_logits)
 
         last_loss = loss
@@ -143,28 +154,24 @@ while current_epoch < MAX_EPOCHS:
         scheduler.step()
 
         if i < SENTENCE_LENGTH - 1:
-            decoder_list[:,i+1] = target_logits[:,i]
-
-
-
-
-
+            decoder_list[:, i + 1] = target_logits[:, i]
 
     current_epoch += 1
 
+    end = time.time_ns()
+
     if current_epoch % 1 == 0:
-        print(f"EPOCH {current_epoch}\n\tLoss: {last_loss}")
-
-        for encoded_sentence, expected_sentence in zip(
-            Transformer.tensor2token(last_prediction[:,:], END_TOKEN), # type: ignore
-            Transformer.tensor2token(target_logits[:,:], END_TOKEN)
-        ):
-            decoded_sentence = TOKENANO.decode(encoded_sentence)
-            decoded_target = TOKENANO.decode(expected_sentence)
-            print(f"\tACTUAL:\n\t\t{decoded_sentence}\n\tEXPECTED:\n\t\t{decoded_target}\n")
-
-
-
-
-
-
+        MIN_LOSS = min(LOSS_HISTORY)
+        MAX_LOSS = max(LOSS_HISTORY)
+        AVERAGE_LOSS = sum(LOSS_HISTORY)/len(LOSS_HISTORY)
+        print(f"EPOCH {current_epoch}\n\tTime: {(end-start)/1E9}s\n\tLoss: {last_loss}")
+        print(f"\tMin Loss: {MIN_LOSS}\tAvg Loss: {AVERAGE_LOSS}\tMax Loss: {MAX_LOSS}\n")
+        # for encoded_sentence, expected_sentence in zip(
+        #     Transformer.tensor2token(last_prediction[:, :], END_TOKEN),  # type: ignore
+        #     Transformer.tensor2token(target_logits[:, :], END_TOKEN),
+        # ):
+        #     decoded_sentence = TOKENANO.decode(encoded_sentence)
+        #     decoded_target = TOKENANO.decode(expected_sentence)
+        #     print(
+        #         f"\tACTUAL:\n\t\t{decoded_sentence}\n\tEXPECTED:\n\t\t{decoded_target}\n"
+        #     )