why a nested Serializer run repeatly 4 times

[flyly0755]

models.py
class Hanzi(AbstractBaseModel):
    zi = models.CharField('汉字', max_length=1, unique=True, null=False, blank=False)
    nandu = models.SmallIntegerField("难度", choices=Constant.HANZI_NANDU, null=True, blank=True)
    zhuyin = models.CharField('主音', max_length=10, null=True, blank=True)
    duoyininfo = JSONField('多音字', null=True, blank=True)
    created_at = models.DateTimeField(verbose_name='Created at', auto_now_add=True, null=True)
    modified_at = models.DateTimeField(verbose_name='Modified at', auto_now=True, null=True)


serializers.py
class HanziListSerializer(serializers.ModelSerializer):
    class Meta:
        model = Hanzi
        fields = ['id', 'zi', 'zhuyin', 'zhuyindiao', 'nandu']

class JSONSerializerField(serializers.Field):
    """ Serializer for JSONField -- required to make field writable"""
    def to_representation(self, data):
        print(17, data, type(data))
        json_data = {}
        try:
            json_data = json.loads(data)
        except ValueError as e:
            pass
        finally:
            print(24, json_data, type(json_data))
            return json_data

    def to_internal_value(self, value):
        tmpvalue = json.loads(value)
        return json.dumps(tmpvalue)

class HanziSerializer(serializers.ModelSerializer):
    duoyininfo = JSONSerializerField()

    class Meta:
        model = Hanzi
        fields = "__all__"


views.py
class HanziViewSet(viewsets.ModelViewSet, BaseViewSetMixin):
    queryset = Hanzi.objects.all()
    serializer_class = HanziSerializer

    filter_backends = [SearchFilter, DjangoFilterBackend]
    search_fields = ['zhuyin'']
    filter_fields = ['zi']

    def get_serializer_class(self):
        if self.action == 'list':
            return HanziListSerializer

        return super(HanziViewSet, self).get_serializer_class()
        # return HanziSerializer

    def perform_update(self, serializer):
        user = self.fill_user(serializer, 'update')
        return serializer.save(**user)

    def perform_create(self, serializer):
        user = self.fill_user(serializer, 'create')
        return serializer.save(**user)

send one get api: /api/word/3632/
with debug print log:
17 {"zhe": "\u671d\uff5e", "zhu\u00f3": "\u7a7f\uff5e"} <class 'jsonfield.json.JSONString'>
24 {'zhe': '朝～', 'zhuó': '穿～'} <class 'dict'>
17 {"zhe": "\u671d\uff5e", "zhu\u00f3": "\u7a7f\uff5e"} <class 'jsonfield.json.JSONString'>
24 {'zhe': '朝～', 'zhuó': '穿～'} <class 'dict'>
17 {"zhe": "\u671d\uff5e", "zhu\u00f3": "\u7a7f\uff5e"} <class 'jsonfield.json.JSONString'>
24 {'zhe': '朝～', 'zhuó': '穿～'} <class 'dict'>
17 {"zhe": "\u671d\uff5e", "zhu\u00f3": "\u7a7f\uff5e"} <class 'jsonfield.json.JSONString'>
24 {'zhe': '朝～', 'zhuó': '穿～'} <class 'dict'>
can see code line: print(17, data, type(data)) and print(24, json_data, type(json_data)) respectively run 4 times, why??