Create REPORT4_Modifications_1+2+3 _PiT_Training_Results_in_Colab

REPORT4_Modifications_1+2+3 _PiT_Training_Results_in_Colab

@@ -0,0 +1,81 @@
+In PiT V3.0, I modified the vanilla PiT model in a first undisclosed manner plus a second dislosed manner, plus a third dislosed manner essentially doubling the original Vanilla ViT parameter count, and the training results greatly improved, and the two-ways modified model was consistently ahead of the unmodified Vanilla PiT model.  
+My threee-ways modified V3.0 PiT model exceeded the vanilla and double-modded PiT Val Accuracy by epoch 8 (exceeding the final/highest 94.75% Val Accuracy of the vanilla PiT model at Epoch ____).
+By Epoch 15, the three-way modified PiT model had the highest Val Accuracy 96.15% (aurpassing the Vanilla PiT models)
+By Epoch 15, the three-ways modified PiT model exceeded 96% Val Accuracy, 
+And Val Accuracy increased up to the highest Val Accuracy of 96.70% (exceeding the Vanilla PiT and the Mod1+Mod2 PiT models before the training was hardcoded terminated Epoch 25.
+
+--- Configuration V3.0  ---
+train_file: /content/sample_data/mnist_train_small.csv
+test_file: /content/sample_data/mnist_test.csv
+image_size: 28
+num_classes: 10
+embed_dim: XXX
+num_layers: X
+num_heads: X
+mlp_dim: XXXX
+dropout: 0.1
+batch_size: 128
+epochs: 25
+learning_rate: 0.0001
+XXXX
+device: cuda
+image_height: 28
+image_width: 28
+sequence_length: 784
+------------------------------------------------------
+
+Data loaded. Training on cuda.
+  Training samples:   17999
+  Validation samples: 2000
+  Test samples:       9999
+
+Model V3.0 initialized with 2,7XX,XXX trainable parameters.
+
+--- Starting Training (V3.0 Model) ---
+Epoch 01/25 | Train Loss: 2.1971 | Val Loss: 1.9749 | Val Acc: 24.40%
+  -> New best validation accuracy! Saving model state.
+Epoch 02/25 | Train Loss: 1.6013 | Val Loss: 0.9542 | Val Acc: 68.20%
+  -> New best validation accuracy! Saving model state.
+Epoch 03/25 | Train Loss: 0.8384 | Val Loss: 0.6123 | Val Acc: 79.80%
+  -> New best validation accuracy! Saving model state.
+Epoch 04/25 | Train Loss: 0.5816 | Val Loss: 0.4429 | Val Acc: 86.30%
+  -> New best validation accuracy! Saving model state.
+Epoch 05/25 | Train Loss: 0.4547 | Val Loss: 0.3766 | Val Acc: 88.70%
+  -> New best validation accuracy! Saving model state.
+Epoch 06/25 | Train Loss: 0.3752 | Val Loss: 0.2995 | Val Acc: 90.90%
+  -> New best validation accuracy! Saving model state.
+Epoch 07/25 | Train Loss: 0.3099 | Val Loss: 0.2500 | Val Acc: 92.45%
+  -> New best validation accuracy! Saving model state.
+Epoch 08/25 | Train Loss: 0.2781 | Val Loss: 0.2305 | Val Acc: 92.95%
+  -> New best validation accuracy! Saving model state.
+Epoch 09/25 | Train Loss: 0.2576 | Val Loss: 0.2144 | Val Acc: 93.25%
+  -> New best validation accuracy! Saving model state.
+Epoch 10/25 | Train Loss: 0.2275 | Val Loss: 0.1862 | Val Acc: 94.45%
+  -> New best validation accuracy! Saving model state.
+Epoch 11/25 | Train Loss: 0.2092 | Val Loss: 0.1714 | Val Acc: 94.60%
+  -> New best validation accuracy! Saving model state.
+Epoch 12/25 | Train Loss: 0.1924 | Val Loss: 0.1534 | Val Acc: 95.10%
+  -> New best validation accuracy! Saving model state.
+Epoch 13/25 | Train Loss: 0.1847 | Val Loss: 0.1468 | Val Acc: 95.40%
+  -> New best validation accuracy! Saving model state.
+Epoch 14/25 | Train Loss: 0.1649 | Val Loss: 0.1480 | Val Acc: 95.30%
+Epoch 15/25 | Train Loss: 0.1587 | Val Loss: 0.1311 | Val Acc: 96.15%
+  -> New best validation accuracy! Saving model state.
+Epoch 16/25 | Train Loss: 0.1503 | Val Loss: 0.1386 | Val Acc: 95.55%
+Epoch 17/25 | Train Loss: 0.1433 | Val Loss: 0.1254 | Val Acc: 95.80%
+Epoch 18/25 | Train Loss: 0.1324 | Val Loss: 0.1160 | Val Acc: 96.05%
+Epoch 19/25 | Train Loss: 0.1285 | Val Loss: 0.1164 | Val Acc: 96.15%
+Epoch 20/25 | Train Loss: 0.1184 | Val Loss: 0.1193 | Val Acc: 96.15%
+Epoch 21/25 | Train Loss: 0.1145 | Val Loss: 0.1123 | Val Acc: 96.45%
+  -> New best validation accuracy! Saving model state.
+Epoch 22/25 | Train Loss: 0.1125 | Val Loss: 0.1200 | Val Acc: 96.05%
+Epoch 23/25 | Train Loss: 0.1053 | Val Loss: 0.1358 | Val Acc: 96.35%
+Epoch 24/25 | Train Loss: 0.1021 | Val Loss: 0.1200 | Val Acc: 96.35%
+Epoch 25/25 | Train Loss: 0.0964 | Val Loss: 0.1061 | Val Acc: 96.70%
+  -> New best validation accuracy! Saving model state.
+--- Training Finished ---
+
+--- Evaluating on Test Set (V3.0 Model) ---
+                                                                   Final Test Loss: 0.1031
+Final Test Accuracy: 96.99%
+-------------------------------------------