def step(self, gradients: List[Optional[Tensor]]):
params = self.param_group['params']
params_with_grad = []
grads = []
state_sums = []
state_steps: List[Tensor] = []
if len(params) != len(gradients):
raise ValueError(
"the gradients passed in does not equal to the size of the parameters!"
+ f"Params length: {len(params)}. " +
f"Gradients length: {len(gradients)}")
has_sparse_grad = False
for param, gradient in zip(self.param_group['params'], gradients):
if gradient is not None:
if gradient.is_sparse:
has_sparse_grad = True
params_with_grad.append(param)
grads.append(gradient)
state = self.state[param]
state_sums.append(state['sum'])
state_steps.append(state['step'])
with torch.no_grad():
F.adagrad(params,
grads,
state_sums,
state_steps,
lr=self.defaults['lr'],
weight_decay=self.defaults['weight_decay'],
lr_decay=self.defaults['lr_decay'],
eps=self.defaults['eps'],
has_sparse_grad=has_sparse_grad,
foreach=self.foreach)
def step(self, gradients: List[Optional[Tensor]]):
params = self.param_group['params']
params_with_grad = []
grads = []
state_sums = []
state_steps: List[int] = []
if len(params) != len(gradients):
raise ValueError(
"the gradients passed in does not equal to the size of the parameters!"
+ f"Params length: {len(params)}. " +
f"Gradients length: {len(gradients)}")
for param, gradient in zip(self.param_group['params'], gradients):
if gradient is not None:
params_with_grad.append(param)
grads.append(gradient)
state = self.state[param]
state_sums.append(state['sum'])
# update the steps for each param group update
state['step'] += 1
# record the step after step update
state_steps.append(state['step'].item())
with torch.no_grad():
F.adagrad(params,
grads,
state_sums,
state_steps,
lr=self.defaults['lr'],
weight_decay=self.defaults['weight_decay'],
lr_decay=self.defaults['lr_decay'],
eps=self.defaults['eps'])