search

FedML-AI / FedML

FEDML - The unified and scalable ML library for large-scale distributed training, model serving, and federated learning. FEDML Launch, a cross-cloud scheduler, further enables running any AI jobs on any GPU cloud or on-premise cluster. Built on this library, TensorOpera AI (https://TensorOpera.ai) is your generative AI platform at scale.
https://TensorOpera.ai
Apache License 2.0
4.2k stars 788 forks source link

Weird accuracy while running fedml-iot with default cnn on MNIST #117

Closed MichaelLee-ceo closed 2 years ago

MichaelLee-ceo commented 3 years ago

It's normal while training under the linear regression on MNIST, but when I change the model to CNN which is located in /fedml_api/model/cv/cnn.py it trains weirdly, are there any solutions for that? Thanks!

chaoyanghe commented 3 years ago

what's the issue you met?

MichaelLee-ceo commented 3 years ago

W B Chart 2021_5_11 上午10_28_07 I just get my training result as this, though I supposed it to be some higher or smoother accuracy. Or how do I need to do to execute the cnn model correctly on fedml-iot?

MichaelLee-ceo commented 3 years ago

How are the steps to execute the cnn version correctly?

chaoyanghe commented 3 years ago

@MichaelLee-ceo you need to tune the hyper-parameters to make it work. Here are some references: https://github.com/FedML-AI/FedML/tree/master/benchmark

Could you share the path and the script you run the experiment?