GitLab

fix max depth

fix maxdepth

* wip_example

* [wip]mnist_pipe_example

* [wip]mnist_pipe_example

* [wip]mnist_pipe_example

* [wip]mnist_pipe_example

* [wip]mnist_oss_example

* working prototype

* added tutorial script

* update tutorial

* Update mnist_test_oss.py

* Update mnist_test_oss.py

* Update mnist_test_oss.py

* Update mnist_test_pipe.py

* Update tutorial_oss.py

* Update tutorial_pipe.py

* Update tutorial_pipe.py

* Update mnist_test_oss.py

* Update tutorial_pipe.py

* Update mnist_test_pipe.py

* Update tutorial_pipe.py

* fix black

* fix flacke8

* general fixes

* add example oss+pipe

* fix isort

* Update mnist_test_pipe.py

* fix black
Co-authored-by: Vittorio Caggiano <caggiano@devfair0253.h2.fair>

* Minor, ease of life to debug and makes it possible to test a host of models with the same code

* small refactor, code cleanup
* broadcast tensor .data attribute directly

- fixed typing
- make it run less often to reduce CI time

testing: run it in a loop make sure it is run in the right frequency.

- close #143

* fixing the readme for oss

* adding a cpu option
* adjust the reference loss

* [fix] fixing circleCI for AdaScale

- ran black, isort, flake8, mypy

* more fix

* Add implementation of AdaScale

* add adascale docs

The expert annotation is used by clip_grads and DDP.

* fixing the issue wrt Apex, validated with Latte, Classy would need another pass

* bugfix
* adjust default non-regression loss, not all_reduced now

* wrapping the model in DDP in the tutorial

* typo

More realistic benchmarks, comparing apples to apples. DDP/OSS+DDP/OSS+SDP

* new unit test to catch rank issues in OSS

Currently only implemented for a single process and expert.

authored-by: Naman Goyal <namangoyal@learnfair0755.h2.fair>

* Add unittest for checkpoint & DDP

- this change adds test cases to reproduce the error with checkpoint & DDP
- mandeep mentioned that there is also deadlock in this case, but this
  change doesn't cover that.
- we cover cases where weight sharing is OK
- however, same module multiple checkpoint or find_unused_parameters are
  both not OK

* added norm checks

Same bucketing strategy for OSS and SDP:
sort everything ahead of time, per rank and per size, smaller tensors first. Bucket the smallest elements in a fixed buffer, send async, then send all the others async, and get back to the bucket. Once done then scatter the contents if needed

support optimizer state sharding for megatron

* minor, but gives some memory back
* adjust CI and regression checks to 4 gpu

- adding the buffer broadcast option
- minor cleanup in shardedDDP

add badges and link to readthedoc

- small benchmark refactor, only one for all backends and ddp
- deterministic, enforce alignment with pytorch ddp

* various fixes, no more issues with `make html` and more API fields should be populated

* Broadcasting grad-enabled tensors is forbidden in Gloo, because this is not differentiable. Workaround