How to let the client train using the initial model provided by the server?

[zhangtian-3841]

What is your question?

Dear Flower developers,

Hello! Thank you for your excellent work in helping me solve many problems! However, I have some questions and I sincerely ask for your assistance:

question 1:

• In the process of implementing federated learning based on Flower and PyTorch, I found that the initial model loaded by the client before training is local.
• However, in certain cases, I need the client to train using the initial model provided by the server(For example, I need the client to continue training using the global model obtained from previous training sessions.).

question 2:

• During the training process, on certain conditions (e.g., a specific port is accessed and requests an immediate halt to the training), how can I make the server or client stop the training process proactively?
• I noticed that the disconnect_all_clients function can stop the federated learning process for a server that exceeds the specified number of rounds. Can I modify it to achieve my goal?

Thank you for your assistance!

[rlakinie]

For question 1, what I did to work was pass the model parameters (provided from the server) into the the strategy definition.

Related: https://flower.ai/docs/framework/tutorial-series-use-a-federated-learning-strategy-pytorch.html