Custom PretrainedModel does not loaded correctly

Beginners
1 
class CustomClassifier(nn.Module):
        .....


class MultipleClassifiers(PreTrainedModel):
        def __init__(self, encoder:AutoModel, config: AutoConfig) -> None:
                super().__init__(config)
                self.encoder = encoder
                self.output_heads = nn.ModuleDict()
                classifier1 = CustomClassfier(encoder, num_classes=100)
                classifier2 = CustomClassfier(encoder, num_classes=1000)
        ...

config = AutoConfig.from_pretrained("distilbert-base")
encoder = AutoModel("distilbert-base")
model = MultipleClassifiers(encoder, config)

# I trained the model using Trainer.
# The problem is when I wan to load the trained model:
model = AutoModel.from_pretrained("path") # load RobertaModel not MultipleClassifiers
model = AutoModelForSequenceClassification.from_pretrained("path") # RobertaForSequenceClassification

How to modify the config or anything else to force the model loading MultipleClassifiers AutoModelForSequenceClassification because I want to use it with TextClassificationPipeline