DL-Art-School/codes/models/steps/steps.py

136 lines
6.0 KiB
Python
Raw Normal View History

2020-08-22 14:24:34 +00:00
from utils.loss_accumulator import LossAccumulator
from torch.nn import Module
import logging
from models.steps.losses import create_generator_loss
import torch
from apex import amp
from collections import OrderedDict
from .injectors import create_injector
2020-09-06 23:27:08 +00:00
from models.novograd import NovoGrad
2020-08-12 14:45:23 +00:00
2020-08-22 14:24:34 +00:00
logger = logging.getLogger('base')
2020-08-12 14:45:23 +00:00
2020-08-22 14:24:34 +00:00
# Defines the expected API for a single training step
class ConfigurableStep(Module):
def __init__(self, opt_step, env):
super(ConfigurableStep, self).__init__()
self.step_opt = opt_step
self.env = env
self.opt = env['opt']
self.gen_outputs = opt_step['generator_outputs']
self.loss_accumulator = LossAccumulator()
self.optimizers = None
2020-08-22 14:24:34 +00:00
self.injectors = []
if 'injectors' in self.step_opt.keys():
for inj_name, injector in self.step_opt['injectors'].items():
self.injectors.append(create_injector(injector, env))
losses = []
self.weights = {}
if 'losses' in self.step_opt.keys():
for loss_name, loss in self.step_opt['losses'].items():
losses.append((loss_name, create_generator_loss(loss, env)))
self.weights[loss_name] = loss['weight']
2020-08-22 14:24:34 +00:00
self.losses = OrderedDict(losses)
# Subclasses should override this to define individual optimizers. They should all go into self.optimizers.
# This default implementation defines a single optimizer for all Generator parameters.
# Must be called after networks are initialized and wrapped.
2020-08-22 14:24:34 +00:00
def define_optimizers(self):
self.training_net = self.env['generators'][self.step_opt['training']] \
if self.step_opt['training'] in self.env['generators'].keys() \
else self.env['discriminators'][self.step_opt['training']]
2020-08-22 14:24:34 +00:00
optim_params = []
for k, v in self.training_net.named_parameters(): # can optimize for a part of the model
if v.requires_grad:
optim_params.append(v)
else:
if self.env['rank'] <= 0:
logger.warning('Params [{:s}] will not optimize.'.format(k))
if 'optimizer' not in self.step_opt.keys() or self.step_opt['optimizer'] == 'adam':
opt = torch.optim.Adam(optim_params, lr=self.step_opt['lr'],
weight_decay=self.step_opt['weight_decay'],
betas=(self.step_opt['beta1'], self.step_opt['beta2']))
elif self.step_opt['optimizer'] == 'novograd':
2020-09-06 23:27:08 +00:00
opt = NovoGrad(optim_params, lr=self.step_opt['lr'], weight_decay=self.step_opt['weight_decay'],
betas=(self.step_opt['beta1'], self.step_opt['beta2']))
2020-08-22 14:24:34 +00:00
self.optimizers = [opt]
2020-08-12 14:45:23 +00:00
# Returns all optimizers used in this step.
def get_optimizers(self):
2020-08-22 14:24:34 +00:00
assert self.optimizers is not None
return self.optimizers
2020-08-12 14:45:23 +00:00
# Returns optimizers which are opting in for default LR scheduling.
def get_optimizers_with_default_scheduler(self):
2020-08-22 14:24:34 +00:00
assert self.optimizers is not None
return self.optimizers
2020-08-12 14:45:23 +00:00
# Returns the names of the networks this step will train. Other networks will be frozen.
def get_networks_trained(self):
2020-08-22 14:24:34 +00:00
return [self.step_opt['training']]
# Performs all forward and backward passes for this step given an input state. All input states are lists of
# chunked tensors. Use grad_accum_step to dereference these steps. Should return a dict of tensors that later
# steps might use. These tensors are automatically detached and accumulated into chunks.
def do_forward_backward(self, state, grad_accum_step, amp_loss_id, train=True):
2020-08-22 14:24:34 +00:00
new_state = {}
2020-08-12 14:45:23 +00:00
2020-08-22 14:24:34 +00:00
# Prepare a de-chunked state dict which will be used for the injectors & losses.
local_state = {}
for k, v in state.items():
local_state[k] = v[grad_accum_step]
local_state.update(new_state)
2020-08-12 14:45:23 +00:00
2020-08-22 14:24:34 +00:00
# Inject in any extra dependencies.
for inj in self.injectors:
# Don't do injections tagged with eval unless we are not in train mode.
if train and 'eval' in inj.opt.keys() and inj.opt['eval']:
continue
2020-08-22 14:24:34 +00:00
injected = inj(local_state)
local_state.update(injected)
new_state.update(injected)
2020-09-09 02:36:27 +00:00
if train and len(self.losses) > 0:
2020-08-22 19:08:33 +00:00
# Finally, compute the losses.
total_loss = 0
for loss_name, loss in self.losses.items():
2020-09-09 02:36:27 +00:00
# Some losses only activate after a set number of steps. For example, proto-discriminator losses can
# be very disruptive to a generator.
if 'after' in loss.opt.keys() and loss.opt['after'] > self.env['step']:
continue
2020-08-22 19:08:33 +00:00
l = loss(self.training_net, local_state)
total_loss += l * self.weights[loss_name]
# Record metrics.
self.loss_accumulator.add_loss(loss_name, l)
for n, v in loss.extra_metrics():
self.loss_accumulator.add_loss("%s_%s" % (loss_name, n), v)
self.loss_accumulator.add_loss("%s_total" % (self.step_opt['training'],), total_loss)
# Scale the loss down by the accumulation factor.
total_loss = total_loss / self.env['mega_batch_factor']
2020-08-22 14:24:34 +00:00
2020-08-22 19:08:33 +00:00
# Get dem grads!
with amp.scale_loss(total_loss, self.optimizers, amp_loss_id) as scaled_loss:
scaled_loss.backward()
2020-08-22 14:24:34 +00:00
# Detach all state variables. Within the step, gradients can flow. Once these variables leave the step
# we must release the gradients.
for k, v in new_state.items():
if isinstance(v, torch.Tensor):
new_state[k] = v.detach()
2020-08-22 14:24:34 +00:00
return new_state
# Performs the optimizer step after all gradient accumulation is completed. Default implementation simply steps()
# all self.optimizers.
2020-08-12 14:45:23 +00:00
def do_step(self):
2020-08-22 14:24:34 +00:00
for opt in self.optimizers:
opt.step()
def get_metrics(self):
2020-08-22 19:08:33 +00:00
return self.loss_accumulator.as_dict()