What is the difference between treemap and opt?

[marcuswang6]

For params 'p', use sample x and y to update it, what's the difference between following codes?
grads = jax.grad(loss)(p, x, y)
updates, state = opt.update(grads, state)
p = optax.apply_updates(p, updates)

grads = jax.grad(loss)(p, x, y)
inner_sgd_fn = lambda g, state: (state - alpha*g)
p = jax.tree_util.tree_multimap(inner_sgd_fn, grads, p)

[YouJiacheng]

optax internally use treemap, but provides a lot of gradient transform to implement advanced optimizer beyond SGD.