Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Communication efficiency optimization #19

Merged
merged 2 commits into from
Apr 22, 2022
Merged

Communication efficiency optimization #19

merged 2 commits into from
Apr 22, 2022

Conversation

xieyxclack
Copy link
Collaborator

  1. The messages are automatically transformed according to their type (e.g., int, float, str) before being fed into the gRPC communicator, which reduces 20%-50% of the message sizes when running toy example in distributed mode. In the previous version, messages are transformed to json_str first and then fed into the gRPC communicator. For more details please refer to:

    • A new proto file gRPC_comm_manager.proto;
    • The transform function and parse function in message.py;

  2. Minor fix for running with distributed mode.

@xieyxclack xieyxclack added the enhancement New feature or request label Apr 15, 2022
@xieyxclack xieyxclack requested review from joneswong and DavdGao April 15, 2022 07:49
@DavdGao
Copy link
Collaborator

DavdGao commented Apr 18, 2022

It looks good to me. (Maybe we can add a unit test for the distributed mode?)

@xieyxclack xieyxclack force-pushed the feature/comm_efficiency branch from 606b11d to 9b0b0bc Compare April 20, 2022 11:29
DavdGao
DavdGao reviewed Apr 20, 2022
View reviewed changes
Copy link
Collaborator

@DavdGao DavdGao left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

looks good to me

joneswong
joneswong approved these changes Apr 22, 2022
View reviewed changes
Copy link
Collaborator

@joneswong joneswong left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The scope of supported types are any composition of dict/list/basic types (int, float, str). It is consistent with what currently being considered in our FL algos.

Approve!

@joneswong joneswong merged commit d5adb81 into master Apr 22, 2022
@joneswong joneswong deleted the feature/comm_efficiency branch April 22, 2022 06:55
cheneydon pushed a commit to cheneydon/FederatedScope that referenced this pull request Jun 14, 2022
@xieyxclack @cheneydon
add README and example_config (#1)
dd5e50f 
Co-authored-by: yuexiang.xyx <[email protected]>

Update `tests/run.py` for Jenkins server (alibaba#4)

just a workaround

Feature/synchronize (alibaba#3)

sync with the master branch of our original gitlab

Feature/config refactor (alibaba#5)

refactored configuration-related code

modify README; minor fix (alibaba#6)

Updated README

fix gan cra loss_batch-> loss_task bug

improved the environments set-up guidance

improved the environments set-up guidance

improved the environments set-up guidance

Fix setup requirements.

Update required python version to 3.9.

updated auto-doc component according to the latest changes

[Feature] Add dropout and log training metric. (alibaba#11)

* Add dropout option for CNN and NLP model; Add training metric to logs.

* allow users to determine whether to conduct evaluation on a specific split

* Enable metric in global eval for users to determine whether to conduct evaluation on a specific split.

* fix minor bug when importing nlp loss

* Replace and remove `validate` with `evaluate(target_data_split_name=split)` to keep code clean.

enabled the log file name valid in windows environment (alibaba#13)

* enabled the log file name valid in windows environment

update readme (alibaba#15)

* update README

added a demo for black-box optimization (alibaba#14)

- added a demo for black-box optimization
- enabled installation with cuda10

[Bugfix] fixed the invalid logger set-up if the logging is used before we call setup_logger (alibaba#17)

* fixed the invalid logger set-up if the `logging` is used before we call `setup_logger`

Change source of `download_url` from our own and fix `README` (alibaba#20)

* Change source of `download_url` from our own `utils.py` and fix `README.md`.

add logo (alibaba#26)

- add logo
- add more icons

modify grpc_comm according to official tutorial (alibaba#25)

fix path issue

fix wrong logger usage

reformatted

Communication efficiency optimization (alibaba#19)

* minor fixed for distributed mode

* For the communication efficiency: dynamic type selection in gRPC servicer; transformer & parser

Refactored the logger by reducing its redundancy and fixed some minor issues (alibaba#29)

* Reducing the redundancy of the logger

Update test_mf.py

modify the unit test of mf task

Refactor splitter&transform; Modify some data related config; Add external dataset. (alibaba#33)

[Feature] FedEx (alibaba#37)

[Feature] FedEx (alibaba#37)

[Hotfix] print the missing ``Final`` results (alibaba#41)

* hotfix for the missing ``Final`` results print

Add pre-trained transformers as NLP model.

TODO:@ZHEN, please fix online aggregator when the device is not specific.

Add a example for transformers.

Fix url. (alibaba#46)

- added the local training baseline
- enabled each client has its own early-stopper

formatted by linter

formatted by linter

not use early_stopper in non-local mode

bugfix for the cast "sample_client_num = -1"

added global training mode via a proxy client that holds all data

Fix un-consistent device for the PIA test

added local fine-tuning before local evaluation

linter format

bugfix for fedex

update README (alibaba#49)

Feature/attack doc (alibaba#50)

* improved the doc for attack module

added API comments (alibaba#52)

Fix docs about graph. (alibaba#51)

Add api ref for mf task and context (alibaba#53)

* add mf api reference and modify README.md

typos fix

Fix minor bugs

Timeout strategy and minimal received number (alibaba#36)

* For async: timeout strategy and minimal received number

modify api reference (alibaba#56)

update doc of core (alibaba#57)

Add datasets from hugging face.

Formatted and fix minor bugs.

Add datasets and scripts for openml.

Modify the example `yaml` of openml datasets.

Add materials (paper lists, tutorials) (alibaba#60)

* add FL paper list

Add paper lists (alibaba#61)

* add FL paper list

fixed some missing API reference in fs.core (alibaba#54)

As the title says.

update release version (alibaba#64)

Update graph paper list. (alibaba#65)

Add paper list for FedHPO (alibaba#67)

* added paper list for fedhpo

rename and modify some val

Add paper list for FedRec (alibaba#68)

add paper list for FedRec

added pfl paper list (alibaba#72)

added pfl paper list

hotfix for transformers to avoid import error

updated pfl paper list (alibaba#73)

updated pfl paper list

fix url in dblp_new.py (alibaba#76)

update README

update

debug squad model

update

update

update
AnthonyXuan pushed a commit to AnthonyXuan/FederatedScope that referenced this pull request Aug 10, 2023
@xieyxclack
Communication efficiency optimization (alibaba#19)
c015cbf 
* minor fixed for distributed mode

* For the communication efficiency: dynamic type selection in gRPC servicer; transformer & parser
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@DavdGao DavdGao DavdGao left review comments

@joneswong joneswong joneswong approved these changes

Assignees
No one assigned
Labels
enhancement New feature or request
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@xieyxclack @DavdGao @joneswong