楼主您好，我很仔细地看过了perfedavg这个代码，但是给我的感觉就是local train 和fine tune都是fedavg的过程，换了一个优化器，就像是进行了两次fedavg，我并没有理解哪里体现了元学习，能否给予解答呢？谢谢！

跟着 示例 走, 当运行 standalone.py 时报错

Traceback (most recent call last):
  File "standalone.py", line 21, in <module>
    from models.cnn import CNN_MNIST
ModuleNotFoundError: No module named 'models'

Hello，
Thanks for reading~

I follow the tutorial from ：https://github.com/SMILELab-FL/FedLab-benchmarks/blob/master/fedlab_benchmarks/datasets/fmnist/fmnist_tutorial.ipynb

However, the following error exists:
FileNotFoundError: [Errno 2] No such file or directory: '../partition-reports/fmnist_noniid-label_1_clients_10.csv'

How can I fix it? Please.

When I try to run the demo, there seems that some files are lost, like the following
ModuleNotFoundError: No module named 'leaf.read_util'.

Moreover, It is suggested that your example on LEAF can be completed in a folder instead of using ''...'' other ways to call the files under the parent folder, it makes the program look messy and prone to the following errors:
ValueError: attempted relative import beyond top-level package

Anyway, this is a nice work for the federated learning experiment. I hope it can be improved.

在使用celeba数据集时，preprocess后 bash gen_pickle_dataset.sh "celeba" "../datasets" "./pickle_datasets"时会报错
FileNotFoundError: [Errno 2] No such file or directory: '/home/fedlab/leaf/datasets/celeba/data/raw/img_align_celeba/162887.jpg'，重新下载数据集或者重新处理，错误仍存在，请问是自己代码的问题吗

报错显示找不到需要的 .pkl 文件。 会不会是在之前的整合目录时没有修改这里。 （我已经在 fedlab_benchmarks/datasets/ 的 /femnist 和 /shakespeare 目录中运行过你们的 README 文档中的 preprocess.sh 示例）

I got this error when executing FedLab-benchmarks/fedlab_benchmarks/fedasync/standalone/standalone.py with fedlab==1.1.2.

Traceback (most recent call last):
  File "standalone.py", line 300, in <module>
    params_list = trainer.train(model_parameters=model_parameters,
  File "/home/xuan/.local/lib/python3.8/site-packages/fedlab/core/client/scale/trainer.py", line 77, in train
    self._LOGGER.info(
AttributeError: 'NoneType' object has no attribute 'info'

Am I missing anything?
Furthermore, which version of fedlab should I used in fedlab_benchmark?

新手想尝试一下联邦学习个性化方向，求教

server manager 传输的 torch.Tensor(self._handler.server_time)
client 回传时直接使用self.model_time + 1
然后server再次拿到model_time进行聚合的时候，staleness = self.current_time - receive_model_time

最后导致传输的model_time tensor持续变大！
2022-04-14 05:34:07,112 - server - INFO - model time : tensor([-1.1402e+13, 1.0000e+00, 1.0000e+00, 1.0000e+00]) 2022-04-14 05:34:07,112 - server - INFO - staleness : tensor([1.1402e+13, 4.0000e+00, 4.0000e+00, 4.0000e+00])

在执行跨进程的异步FL任务时会陷入等待。

hm@hm-IdeaCentre-GeekPro-15ICK:~/test/fedasync$ bash quick_start.sh 
torch.distributed connection is initializing with ip address 127.0.0.1:3002, rank 0, world size: 3, backend gloo, ethernet None.
torch.distributed connection is initializing with ip address 127.0.0.1:3002, rank 1, world size: 3, backend gloo, ethernet None.
torch.distributed connection is initializing with ip address 127.0.0.1:3002, rank 2, world size: 3, backend gloo, ethernet None.
2022-04-07 22:13:36,811 - root - INFO - Added key: store_based_barrier_key:1 to store for rank: 1
2022-04-07 22:13:36,811 - root - INFO - Added key: store_based_barrier_key:1 to store for rank: 0
2022-04-07 22:13:36,811 - root - INFO - Added key: store_based_barrier_key:1 to store for rank: 2
2022-04-07 22:13:36,812 - client 1 - INFO - request parameter procedure
2022-04-07 22:13:36,812 - client 2 - INFO - request parameter procedure