def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
self.scores = nn.ParameterList([
nn.Parameter(module_util.mask_init(self))
for _ in range(pargs.num_seed_tasks_learned)
])
for s in self.scores:
s.requires_grad = False
self.scores.requires_grad = False
if pargs.train_weight_tasks == 0:
self.weight.requires_grad = False
if pargs.start_at_optimal:
self.basis_alphas = nn.ParameterList([
nn.Parameter(torch.eye(pargs.num_seed_tasks_learned)[i])
for i in range(pargs.num_seed_tasks_learned)
] + [
nn.Parameter(
torch.ones(pargs.num_seed_tasks_learned) /
pargs.num_seed_tasks_learned)
for _ in range(pargs.num_seed_tasks_learned, pargs.num_tasks)
])
else:
self.basis_alphas = nn.ParameterList([
nn.Parameter(
torch.ones(pargs.num_seed_tasks_learned) /
pargs.num_seed_tasks_learned)
for _ in range(pargs.num_tasks)
])
self.sparsity = pargs.sparsity
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
# initialize the scores
self.scores = nn.Parameter(module_util.mask_init(self))
# Turn the gradient on the weights off
self.weight.requires_grad = False
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
self.scores = nn.ParameterList([
nn.Parameter(module_util.mask_init(self))
for _ in range(pargs.num_tasks)
])
if pargs.train_weight_tasks == 0:
self.weight.requires_grad = False
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
mask_init = module_util.mask_init(self)
# initialize the scores
d = {set: nn.Parameter(mask_init.clone()) for set in pargs.set}
d['INIT'] = nn.Parameter(mask_init.clone())
self.scores = nn.ParameterDict(d)
# Turn the gradient on the weights off
self.weight.requires_grad = False
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
# initialize the scores
self.scores = nn.Parameter(module_util.mask_init(self))
# Turn the gradient on the weights off
if pargs.train_weight_tasks == 0:
self.weight.requires_grad = False
# default sparsity
self.sparsity = pargs.sparsity