Qing Wang
Qing Wang
The fed_task_id is not the same in different places in current codebase. see: https://github.com/ray-project/rayfed/blob/main/fed/_private/fed_call_holder.py#L61 https://github.com/ray-project/rayfed/blob/main/fed/fed_object.py#L40
for the following actions: - [ ] the abstractions and architectures - [ ] for some local code paths - [ ] the doc-strings and comments - [ ] logging...
There are many duplicate stuff in our test infra codebase. This issue is aiming to refactor that.
In the research and simulation scenarios, users may need to run large workloads within limited resources. If we use raw Ray actors as fed instances, it may limit the workloads...
Add an e2e example, with using ray datatset for dataloading and preparing, and an pytorch NN for federated training.
I'm proposing that support running rayfed job in single-controller mode. I'd like to propose 2 options on how we startup the single-controller cluster and how we connect to the cluster...
Maybe this is one of the micro kernel arch items? #107