QuantizedMatmul¶

class paddle.nn.quant.quant_layers. QuantizedMatmul ( layer: Layer | None = None, weight_bits: int = 8, activation_bits: int = 8, moving_rate: float = 0.9, weight_quantize_type: _QuantType = 'abs_max', activation_quantize_type: _QuantType = 'abs_max', weight_pre_layer: Layer | None = None, act_pre_layer: Layer | None = None, weight_quant_layer: Layer | None = None, act_quant_layer: Layer | None = None ) [source]

The computational logic of QuantizedMatmul is the same with Matmul. The only difference is that its inputs are all fake quantized.

forward ( x: Tensor, y: Tensor, transpose_x: bool = False, transpose_y: bool = False, name: str | None = None ) → Tensor

forward¶

Defines the computation performed at every call. Should be overridden by all subclasses.

Parameters

• *inputs (tuple) – unpacked tuple arguments

• **kwargs (dict) – unpacked dict arguments