megengine.quantization.LSQ

class LSQ(dtype, enable=True, eps=1e-05, **kwargs)

LSQ: https://arxiv.org/pdf/1902.08153.pdf Estimating and scaling the task loss gradient at each weight and activation layer’s quantizer step size

参数

• dtype (Union[str, QuantDtypeMeta]) – a string or QuantDtypeMeta indicating the target quantization dtype of input.

• enable (bool) – whether do normal_forward or fake_quant_forward.

• eps (float) – a small value to avoid division by zero. Default: 1e-5

Methods

apply(fn)	Applies function fn to all the modules within this module, including itself.
buffers([recursive])	Returns an iterable for the buffers of the module.
children(**kwargs)	Returns an iterable for all the submodules that are direct attributes of this module.
disable()
disable_quantize([value])	Sets module's quantize_disabled attribute and return module.
enable()
eval()	Sets training mode of all the modules within this module (including itself) to False.
fake_quant_forward(inp[, qparams])
forward(inp[, qparams])
get_qparams()
get_quantized_dtype()
load_state_dict(state_dict[, strict])	Loads a given dictionary created by state_dict into this module.
modules(**kwargs)	Returns an iterable for all the modules within this module, including itself.
named_buffers([prefix, recursive])	Returns an iterable for key buffer pairs of the module, where key is the dotted path from this module to the buffer.
named_children(**kwargs)	Returns an iterable of key-submodule pairs for all the submodules that are direct attributes of this module, where 'key' is the attribute name of submodules.
named_modules([prefix])	Returns an iterable of key-module pairs for all the modules within this module, including itself, where 'key' is the dotted path from this module to the submodules.
named_parameters([prefix, recursive])	Returns an iterable for key Parameter pairs of the module, where key is the dotted path from this module to the Parameter.
named_tensors([prefix, recursive])	Returns an iterable for key tensor pairs of the module, where key is the dotted path from this module to the tensor.
normal_forward(inp[, qparams])
parameters([recursive])	Returns an iterable for the Parameter of the module.
register_forward_hook(hook)	Registers a hook to handle forward results.
register_forward_pre_hook(hook)	Registers a hook to handle forward inputs.
replace_param(params, start_pos[, seen])	Replaces module's parameters with params, used by ParamPack to
set_qparams(qparams)
state_dict([rst, prefix, keep_var])
tensors([recursive])	Returns an iterable for the Tensor of the module.
train([mode, recursive])	Sets training mode of all the modules within this module (including itself) to mode.
zero_grad()	Sets all parameters' grads to zero

上一页

megengine.quantization.TQT.zero_grad

下一页

megengine.quantization.LSQ.apply