QuantizationFreezePass¶

class paddle.fluid.contrib.slim.quantization.quantization_pass. QuantizationFreezePass ( scope, place, bias_correction=False, weight_bits=8, activation_bits=8, weight_quantize_type='abs_max', quantizable_op_type=None ) [source]

apply ( graph )

apply¶

Adjust quantize/dequantize operators order for the inference process.

Parameters

graph (IrGraph) – the applied graph.

Returns

None