import copy import warnings import abc import torch from torch import nn from torch.nn.utils import parametrize from torch.nn.modules.container import ModuleDict, ModuleList from .parametrization import PruningParametrization, ZeroesParametrization, ActivationReconstruction, BiasHook from torch.ao.sparsity import BaseSparsifier, module_to_fqn, fqn_to_module SUPPORTED_MODULES = { # added to config if None given nn.Linear, nn.Conv2d, nn.BatchNorm2d, # will need manual update to match conv2d } NEEDS_ZEROS = { # these layers should have pruned indices zero-ed, not removed nn.BatchNorm2d } class BasePruner(BaseSparsifier): r"""Base class for all pruners. Abstract methods that need to be implemented: - update_mask: Function to compute a new mask for all keys in the `module_groups`. Args: - defaults [dict]: default configurations will be attached to the configuration. Only the keys that don't exist in the `config` will be updated. - also_prune_bias [bool]: whether to prune bias in addition to weights (to prune full output channel) or not; default=True. """ def __init__(self, defaults, also_prune_bias=True): super().__init__(defaults) self.prune_bias = also_prune_bias def _prepare(self, use_path=False, *args, **kwargs): r"""Adds mask parametrization to the layer weight """ self.activation_handles = [] # store removable hook handles self.bias_handles = [] for config in self.module_groups: modules = [] if use_path: if type(config['module']) is tuple: # (Conv2d, BN) for fqn in config['fqn']: module = fqn_to_module(self.model, fqn) modules.append(module) else: module = fqn_to_module(self.model, config['fqn']) modules.append(module) else: if type(config['module']) is tuple: for module in config['module']: modules.append(module) else: module = config['module'] modules.append(module) for module in modules: if not isinstance(module, tuple(NEEDS_ZEROS)): # add pruning parametrization and forward hooks if getattr(module, 'mask', None) is None: module.register_buffer('mask', torch.tensor(module.weight.shape[0])) param = config.get('parametrization', PruningParametrization) parametrize.register_parametrization(module, 'weight', param(module.mask), unsafe=True) assert isinstance(module.parametrizations, ModuleDict) # make mypy happy assert isinstance(module.parametrizations.weight, ModuleList) if isinstance(module, tuple(SUPPORTED_MODULES)): self.activation_handles.append(module.register_forward_hook( ActivationReconstruction(module.parametrizations.weight[0]) )) else: raise NotImplementedError("This module type is not supported yet.") else: # needs zeros if getattr(module, 'mask', None) is None: module.register_buffer('mask', torch.tensor(module.weight.shape[0])) param = config.get('parametrization', ZeroesParametrization) parametrize.register_parametrization(module, 'weight', param(module.mask), unsafe=True) if module.bias is not None: module.register_parameter('_bias', nn.Parameter(module.bias.detach())) module.bias = None self.bias_handles.append(module.register_forward_hook(BiasHook(module.parametrizations.weight[0], self.prune_bias))) if len(modules) == 2: # (Conv2d, BN) # should have the same set of pruned outputs modules[1].parametrizations.weight[0].pruned_outputs = modules[0].parametrizations.weight[0].pruned_outputs def prepare(self, model, config): r"""Prepares a model, by adding the parametrizations and forward post-hooks. Note:: The model is modified inplace. If you need to preserve the original model, use copy.deepcopy. Args: - model [nn.Module]: model to configure. The model itself is not saved but used for the state_dict saving / loading. - config [list]: configuration elements could either be instances of nn.Module or dict maps. The dicts must have a key 'module' with the value being an instance of a nn.Module. """ self.model = model # TODO: Need to figure out how to load without this. self.config = config # If no config -- try getting all the supported layers if self.config is None: # Add all models to the config self.config = [] stack = [model] while stack: module = stack.pop() for name, child in module.named_children(): if type(child) in SUPPORTED_MODULES: self.config.append(child) else: if type(child) in NEEDS_ZEROS and self.prune_bias: warnings.warn(f"Models with {type(child)} layers have config provided by user.") stack.append(child) for module_config in self.config: if type(module_config) is tuple: first_layer, next_layer = module_config assert isinstance(first_layer, nn.Conv2d) and isinstance(next_layer, nn.BatchNorm2d) module_config = {'module': module_config} local_args = copy.deepcopy(self.defaults) local_args.update(module_config) fqn_list = [] for module in local_args['module']: module_fqn = module_to_fqn(model, module) if module_fqn and module_fqn[0] == '.': module_fqn = module_fqn[1:] fqn_list.append(module_fqn) local_args['fqn'] = fqn_list else: if isinstance(module_config, nn.Module): module_config = {'module': module_config} local_args = copy.deepcopy(self.defaults) local_args.update(module_config) module = local_args['module'] module_fqn = module_to_fqn(model, module) if module_fqn and module_fqn[0] == '.': module_fqn = module_fqn[1:] local_args['fqn'] = module_fqn self.module_groups.append(local_args) self._prepare() def squash_mask(self, use_path=False, *args, **kwargs): for config in self.module_groups: modules = [] if use_path: if type(config['module']) is tuple: # (Conv2d, BN) for fqn in config['fqn']: module = fqn_to_module(self.model, fqn) modules.append(module) else: module = fqn_to_module(self.model, config['fqn']) modules.append(module) else: if type(config['module']) is tuple: for module in config['module']: modules.append(module) else: module = config['module'] modules.append(module) for module in modules: parametrize.remove_parametrizations(module, 'weight', leave_parametrized=True) if getattr(module._parameters, 'mask', None): del module._parameters['mask'] elif getattr(module._buffers, 'mask', None): del module._buffers['mask'] delattr(module, 'mask') def get_module_pruned_outputs(self, module): r"""Returns the set of pruned indices of module""" assert parametrize.is_parametrized(module) # can only get pruned indices of pruned module return module.parametrizations.weight[0].pruned_outputs # assume only one parametrization attached def step(self, use_path=False): if not self.enable_mask_update: return with torch.no_grad(): for config in self.module_groups: modules = [] if use_path: if type(config['module']) is tuple: # (Conv2d, BN) for fqn in config['fqn']: module = fqn_to_module(self.model, fqn) modules.append(module) else: module = fqn_to_module(self.model, config['fqn']) modules.append(module) else: if type(config['module']) is tuple: for module in config['module']: modules.append(module) else: module = config['module'] modules.append(module) # only need to update the first module in modules if len(modules) > 1 # since they should share the same set of pruned outputs module = modules[0] self.update_mask(module, **config) @abc.abstractmethod def update_mask(self, layer, **kwargs): pass