src/create_model.py

from transformers import PretrainedConfig, PreTrainedModel
import torch
import torch.nn as nn


class ONNXBaseConfig(PretrainedConfig):
    model_type = "onnx-base"

    def __init__(self, model_path=None, **kwargs):
        self.model_path = model_path
        super().__init__(**kwargs)


model_directory = './new_model'

config = ONNXBaseConfig(model_path='model.onnx')
config.save_pretrained(save_directory=model_directory)


class ONNXBaseModel(PreTrainedModel):
    config_class = ONNXBaseConfig
    def __init__(self, config):
        super().__init__(config)
        self.dummy_param = nn.Parameter(torch.zeros(0))

    def forward(self, inputs):
        return torch.zeros_like(inputs)

    def save_pretrained(self, save_directory: str, **kwargs):
        super().save_pretrained(save_directory=save_directory, **kwargs)
        onnx_file_path = save_directory + '/model.onnx'
        dummy_input = torch.tensor([[1, 2], [3, 4]], dtype=torch.float32)
        torch.onnx.export(self, dummy_input, onnx_file_path,
                          input_names=['input'], output_names=['output'],
                          dynamic_axes={'input': {0: 'batch_size'}, 'output': {0: 'batch_size'}})


# Initialize model
model = ONNXBaseModel(config)
# Save model
model.save_pretrained(save_directory=model_directory)

model = model.from_pretrained(model_directory)

# Test model
dummy_input = torch.tensor([[1, 2], [3, 4]], dtype=torch.float32)
output_tensor = model(dummy_input)
print(output_tensor)

# Test the onnx model
onnx_file_path = model_directory + '/model.onnx'
import onnx
import onnxruntime as ort

ort_session = ort.InferenceSession(onnx_file_path)
outputs = ort_session.run(None, {'input': dummy_input.numpy()})
print("Model output:", outputs)