possible max_grad_norm implementation error

[robertreaney]

Clipping is currently applied after the optimizer step. The gradient clipping code controlled by the max-grad-norm cli param should be implemented between the backward and step calls.

Node Prediction Code

graphstorm/python/graphstorm/trainer/np_trainer.py

Lines 214 to 221 in f3a0636

	self.optimizer.zero_grad()
	loss.backward()
	rt_profiler.record('train_backward')
	self.optimizer.step()
	rt_profiler.record('train_step')
	if max_grad_norm is not None:
	th.nn.utils.clip_grad_norm_(model.parameters(), max_grad_norm, grad_norm_type)