load model failed with custom torch.nn.Module class 2022-07-18 01:58:04,499 [INFO ] W-9000-model_1-stdout org.pytorch.serve.wlm.WorkerLifeCycle - AttributeError: Can’t get attribute ‘transformer_Model’ on <module ‘main’ from ‘/opt/conda/lib/python3.6/site-packages/ts/model_service_worker.py’>

[bnicholl]

When I am trying to deploy a custom neural network, made like so

import torch.nn as nn
import torch.nn.functional as F

class Model(nn.Module):
    def __init__(self):
        super().__init__()
        self.conv1 = nn.Conv2d(1, 20, 5)
        self.conv2 = nn.Conv2d(20, 20, 5)

    def forward(self, x):
        x = F.relu(self.conv1(x))
        return F.relu(self.conv2(x))

neural_net = Model()

I get the error,
class 2022-07-18 01:58:04,499 [INFO ] W-9000-model_1-stdout org.pytorch.serve.wlm.WorkerLifeCycle - AttributeError: Can’t get attribute ‘transformer_Model’ on <module ‘main’ from ‘/opt/conda/lib/python3.6/site-packages/ts/model_service_worker.py’>

I believe this is because the
type(neural_net) = __main__.Model . So when I use sagemaker's PyTorchModel function to deploy, it is not seeing it as a pytorch object, and thus erroring. This shouldn't be the case though, because the above class is how you are supposed to create custom pytorch classifiers.

To further elaborate how I'm creating my endpoint, in a sagemaker note book I run

val_to_col_elec_model = PyTorchModel(
                         model_data='s3://path_to_model/model_.tar.gz',
                         role=role,
                         entry_point='torchserve.py',
                         source_dir='value_to_column_source_dir',
                         framework_version='1.6.0',
                         py_version='py36')

in order to instantiate the model
in another cell, run

nl_detector = val_to_col_elec_model.deploy(
                     initial_instance_count = 1,
                     serializer=JSONSerializer(),
                     deserializer=JSONDeserializer(),
                     instance_type = 'ml.t2.medium', endpoint_name = endpoint_name)

to deploy.

Just to clarify, I have tested this code and endpoint on various different pytorch models, and they only seem to not work when I create a pytorhc model using the above class Model(nn.Module):