switch beautiful_mnist to use new optimizer [pr] (#8231)

* switch beautiful_mnist to use new optimizer [pr]

* fix abstractions3 + docs

* fix OptimizerGroup with schedule_step api

docs/abstractions3.py

@@ -26,10 +26,11 @@ l1n, l2n = l1.numpy(), l2.numpy()
 from tinygrad.nn.optim import SGD
 optim = SGD([l1, l2])
 
+Tensor.training = True
 X, Y = X_train[(samples:=Tensor.randint(128, high=X_train.shape[0]))], Y_train[samples]
 optim.zero_grad()
 model(X).sparse_categorical_crossentropy(Y).backward()
-optim._step()   # this will step the optimizer without running realize
+optim.schedule_step()   # this will step the optimizer without running realize
 
 # *****
 # 3. Create a schedule.

docs/tensor/properties.md

@@ -31,4 +31,8 @@
 ::: tinygrad.Tensor.shard_
 ::: tinygrad.Tensor.contiguous
 ::: tinygrad.Tensor.contiguous_backward
+
+## Gradient
+
+::: tinygrad.Tensor.gradient
 ::: tinygrad.Tensor.backward