quark.onnx.refine#

Module Contents#

Functions#

quark.onnx.refine.adjust_quantize_info(model: onnx.ModelProto, adjust_shift_cut: bool = True, adjust_shift_bias: bool = True, adjust_shift_read: bool = True, adjust_shift_write: bool = True, adjust_hard_sigmoid: bool = True, adjust_shift_swish: bool = True, align_concat: bool = True, align_pool: bool = True, align_pad: bool = True, align_slice: bool = True) quark.onnx.quant_utils.ONNXQuantizedModel#

Adjust the quantize info to meet the compiler constraints.

quark.onnx.refine.align_quantize_info(model: onnx.ModelProto, align_concat: bool = True, align_pool: bool = True, align_pad: bool = True, align_slice: bool = True, align_transpose: bool = True, align_reshape: bool = True) Any#

Align the quantize info to meet the compiler constraints. This function supports pof2 scale and float scale both