Add debug prints to training test and roll back to 10% ATOL

forge/test/mlir/mnist/training/test_training.py

@@ -146,7 +146,16 @@ def test_mnist_training_with_grad_accumulation():
             total_loss += loss.item()
 
             golden_loss = loss_fn(golden_pred, target)
-            assert torch.allclose(loss, golden_loss, rtol=5e-2)  # 5% tolerance
+
+            print(f"Loss: {loss}")
+            print(f"Golden Loss: {golden_loss}")
+
+            diff = torch.abs(loss - golden_loss)
+            relative_diff = diff / torch.abs(golden_loss)
+            print(f"Absolute Difference: {diff}")
+            print(f"Relative Difference: {relative_diff}")
+
+            assert torch.allclose(loss, golden_loss, rtol=1e-1)  # 10% tolerance
 
             # Run backward pass on device
             loss.backward()