Python torch._utils方法代码示例

# 需要导入模块: import torch [as 别名]
# 或者: from torch import _utils [as 别名]
def _rebuild_tensor_v2(storage, storage_offset, size, stride, requires_grad, backward_hooks):
        tensor = torch._utils._rebuild_tensor(storage, storage_offset, size, stride)
        tensor.requires_grad = requires_grad
        tensor._backward_hooks = backward_hooks
        return tensor 

# 需要导入模块: import torch [as 别名]
# 或者: from torch import _utils [as 别名]
def prep_param_lists(model, flat_master=False):
    """
    Creates a list of FP32 master parameters for a given model, as in
    `Training Neural Networks with Mixed Precision:  Real Examples`_.
    Args:
        model (torch.nn.Module): Existing Pytorch model
        flat_master (bool, optional, default=False):  Flatten the master parameters into a single tensor, as a performance optimization.
    Returns:
        A tuple (``model_params``, ``master_params``). ``model_params`` is a list of the model's parameters for later use with :func:`model_grads_to_master_grads` and :func:`master_params_to_model_params`.  ``master_params`` is a list of FP32 master gradients.  If ``flat_master=True``, ``master_params`` will be a list with one element.
    Example::
        model_params, master_params = prep_param_lists(model)
    .. warning::
        Currently, if ``flat_master=True``, all the model's parameters must be the same type.  If the model has parameters of different types, use ``flat_master=False``, or use :class:`FP16_Optimizer`.
    .. _`Training Neural Networks with Mixed Precision:  Real Examples`:
        http://on-demand.gputechconf.com/gtc/2018/video/S81012/
    """
    model_params = [param for param in model.parameters() if param.requires_grad]

    if flat_master:
        # Give the user some more useful error messages
        try:
            # flatten_dense_tensors returns a contiguous flat array.
            # http://pytorch.org/docs/master/_modules/torch/_utils.html
            master_params = _flatten_dense_tensors([param.data for param in model_params]).float()
        except:
            print("Error in prep_param_lists:  model may contain a mixture of parameters "
                      "of different types.  Use flat_master=False, or use F16_Optimizer.")
            raise
        master_params = torch.nn.Parameter(master_params)
        master_params.requires_grad = True
        # master_params.register_hook(backwards_debug_hook)
        if master_params.grad is None:
            master_params.grad = master_params.new(*master_params.size())
        return model_params, [master_params]
    else:
        master_params = [param.clone().float().detach() for param in model_params]
        for param in master_params:
            param.requires_grad = True
        return model_params, master_params 

# 需要导入模块: import torch [as 别名]
# 或者: from torch import _utils [as 别名]
def prep_param_lists(model, flat_master=False):
    """
    Creates a list of FP32 master parameters for a given model, as in
    `Training Neural Networks with Mixed Precision:  Real Examples`_.

    Args:
        model (torch.nn.Module): Existing Pytorch model
        flat_master (bool, optional, default=False):  Flatten the master parameters into a single tensor, as a performance optimization.
    Returns:
        A tuple (``model_params``, ``master_params``). ``model_params`` is a list of the model's parameters for later use with :func:`model_grads_to_master_grads` and :func:`master_params_to_model_params`.  ``master_params`` is a list of FP32 master gradients.  If ``flat_master=True``, ``master_params`` will be a list with one element.

    Example::

        model_params, master_params = prep_param_lists(model)

    .. warning::
        Currently, if ``flat_master=True``, all the model's parameters must be the same type.  If the model has parameters of different types, use ``flat_master=False``, or use :class:`FP16_Optimizer`.

    .. _`Training Neural Networks with Mixed Precision:  Real Examples`:
        http://on-demand.gputechconf.com/gtc/2018/video/S81012/
    """
    model_params = [param for param in model.parameters() if param.requires_grad]

    if flat_master:
        # Give the user some more useful error messages
        try:
            # flatten_dense_tensors returns a contiguous flat array.
            # http://pytorch.org/docs/master/_modules/torch/_utils.html
            master_params = _flatten_dense_tensors([param.data for param in model_params]).float()
        except:
            print("Error in prep_param_lists:  model may contain a mixture of parameters "
                      "of different types.  Use flat_master=False, or use F16_Optimizer.")
            raise
        master_params = torch.nn.Parameter(master_params)
        master_params.requires_grad = True
        # master_params.register_hook(backwards_debug_hook)
        if master_params.grad is None:
            master_params.grad = master_params.new(*master_params.size())
        return model_params, [master_params]
    else:
        master_params = [param.clone().float().detach() for param in model_params]
        for param in master_params:
            param.requires_grad = True
        return model_params, master_params 

# 需要导入模块: import torch [as 别名]
# 或者: from torch import _utils [as 别名]
def prep_param_lists(model, flat_master=False):
    """
    Creates a list of FP32 master parameters for a given model, as in 
    `Training Neural Networks with Mixed Precision:  Real Examples`_.

    Args:
        model (torch.nn.Module): Existing Pytorch model
        flat_master (bool, optional, default=False):  Flatten the master parameters into a single tensor, as a performance optimization.
    Returns:
        A tuple (``model_params``, ``master_params``). ``model_params`` is a list of the model's parameters for later use with :func:`model_grads_to_master_grads` and :func:`master_params_to_model_params`.  ``master_params`` is a list of FP32 master gradients.  If ``flat_master=True``, ``master_params`` will be a list with one element.

    Example::

        model_params, master_params = prep_param_lists(model)

    .. warning::
        Currently, if ``flat_master=True``, all the model's parameters must be the same type.  If the model has parameters of different types, use ``flat_master=False``, or use :class:`FP16_Optimizer`.

    .. _`Training Neural Networks with Mixed Precision:  Real Examples`:
        http://on-demand.gputechconf.com/gtc/2018/video/S81012/
    """
    model_params = [param for param in model.parameters() if param.requires_grad]

    if flat_master:
        # Give the user some more useful error messages
        try:
            # flatten_dense_tensors returns a contiguous flat array.
            # http://pytorch.org/docs/master/_modules/torch/_utils.html
            master_params = _flatten_dense_tensors([param.data for param in model_params]).float()
        except TypeError as instance:
            # This is brittle, and depends on how cat chooses to word its error message.
            if "cat received an invalid combination of arguments" not in instance.args[0]:
                raise
            else:
                # If you append a message to the exception instance, via
                # instance.args = instance.args + ("Error...",)
                # this messes up the terminal-formatted printing of the instance's original message.
                # Basic solution for now:
                print("Error in prep_param_lists:  model likely contains a mixture of parameters "
                      "of different types.  Use flat_master=False, or use F16_Optimizer.")
                raise
        master_params = torch.nn.Parameter(master_params)
        master_params.requires_grad = True
        # master_params.register_hook(backwards_debug_hook)
        if master_params.grad is None:
            master_params.grad = master_params.new(*master_params.size())
        return model_params, [master_params]
    else:
        master_params = [param.clone().float().detach() for param in model_params]
        for param in master_params:
            param.requires_grad = True
        return model_params, master_params 

# 需要导入模块: import torch [as 别名]
# 或者: from torch import _utils [as 别名]
def prep_param_lists(model, flat_master=False):
    """
    Creates a list of FP32 master parameters for a given model, as in 
    `Training Neural Networks with Mixed Precision:  Real Examples`_.

    Args:
        model (torch.nn.Module): Existing Pytorch model
        flat_master (bool, optional, default=False):  Flatten the master parameters into a single tensor, as a performance optimization.
    Returns:
        A tuple (``model_params``, ``master_params``). ``model_params`` is a list of the model's parameters for later use with :func:`model_grads_to_master_grads` and :func:`master_params_to_model_params`.  ``master_params`` is a list of FP32 master gradients.  If ``flat_master=True``, ``master_params`` will be a list with one element.

    Example::

        model_params, master_params = prep_param_lists(model)

    .. warning::
        Currently, if ``flat_master=True``, all the model's parameters must be the same type.  If the model has parameters of different types, use ``flat_master=False``, or use :class:`FP16_Optimizer`.

    .. _`Training Neural Networks with Mixed Precision:  Real Examples`:
        http://on-demand.gputechconf.com/gtc/2018/video/S81012/
    """
    model_params = [param for param in model.parameters() if param.requires_grad]

    if flat_master:
        # Give the user some more useful error messages
        try:
            # flatten_dense_tensors returns a contiguous flat array.
            # http://pytorch.org/docs/master/_modules/torch/_utils.html
            master_params = _flatten_dense_tensors([param.data for param in model_params]).float()
        except:
            print("Error in prep_param_lists:  model may contain a mixture of parameters "
                      "of different types.  Use flat_master=False, or use F16_Optimizer.")
            raise
        master_params = torch.nn.Parameter(master_params)
        master_params.requires_grad = True
        # master_params.register_hook(backwards_debug_hook)
        if master_params.grad is None:
            master_params.grad = master_params.new(*master_params.size())
        return model_params, [master_params]
    else:
        master_params = [param.clone().float().detach() for param in model_params]
        for param in master_params:
            param.requires_grad = True
        return model_params, master_params