MCPcopy Create free account
hub / github.com/pytorch/executorch / annotate_conv2d

Function annotate_conv2d

backends/samsung/quantizer/annotator.py:651–677  ·  view source on GitHub ↗
(node: Node, quant_config: QuantizationConfig)

Source from the content-addressed store, hash-verified

649 ]
650)
651def annotate_conv2d(node: Node, quant_config: QuantizationConfig) -> None:
652 # skipping quantization if weights are not float
653 if _is_annotated([node]) or not _is_float_tensor(node.args[1]):
654 return
655
656 input = node.args[0]
657 # input & weight (or bias) setting for Conv node(producer_node)
658 quantization_annotation = node.meta.get(
659 "quantization_annotation", QuantizationAnnotation()
660 )
661 if quantization_annotation.input_qspec_map is None:
662 quantization_annotation.input_qspec_map = {}
663
664 shared_quant_node = _get_quantization_annotation(input)
665 if shared_quant_node:
666 quantization_annotation.input_qspec_map[input] = SharedQuantizationSpec(
667 shared_quant_node
668 )
669 else:
670 quantization_annotation.input_qspec_map[input] = quant_config.input_activation
671 quantization_annotation.input_qspec_map[node.args[1]] = quant_config.weight
672 if len(node.args) > 2 and quant_config.bias is not None:
673 quantization_annotation.input_qspec_map[node.args[2]] = quant_config.bias
674 quantization_annotation.output_qspec = quant_config.output_activation
675
676 node.meta["quantization_annotation"] = quantization_annotation
677 node.meta["quantization_annotation"]._annotated = True
678
679
680# CASE 8: embedding

Callers

nothing calls this directly

Calls 4

_is_annotatedFunction · 0.70
_is_float_tensorFunction · 0.70
getMethod · 0.45

Tested by

no test coverage detected