Accessing Model Parameters from tflite file

TensorFlow General Discussion
,
1

Hi

I’m just working on Federated Learning. As my use case demands on-device training of Machine Leaning models, I need to get access to the model parameters for cloud based model aggregation. Is it possible to access the model parameters from the tflite file?

PS: The tflite file is an hexadecimal 1D array as per my understanding

Thanks

3

Hi @Sriram_Srinivasan,
Yes, the model parameters can be accessible from tflite file as the tflite saves the tensor data, model parameters in a structured flatbuffer format. Use tflite_model_analyzer to breakdown the flatbuffer structure for weights and biases and also try netron app for visualizing model architecture. Instead of model parameters extraction from tflite, it might be better to use tensorflow federated framework for your usecase . Thank You