GitLab

* relocate core integration tests

* add sys.path context manager

* cleanup

* try

* try2

* fix path

* doc

* style

* add dep

* add 2 more deps

* synced gpus

* fix

* fix

* need to use t5-small for quality tests

* notes

* complete merge

* fix a disappearing std stream problem

* start zero3 tests

* wip

* tune params

* sorting out the pre-trained model loading

* reworking generate loop wip

* wip

* style

* fix tests

* split the tests

* refactor tests

* wip

* parameterized

* fix

* workout the resume from non-ds checkpoint pass + test

* cleanup

* remove no longer needed code

* split getter/setter functions

* complete the docs

* suggestions

* gpus and their compute capabilities link

* Apply suggestions from code review
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* style

* remove invalid paramgd

* automatically configure zero3 params that rely on hidden size

* make _get_resized_embeddings zero3-aware

* add test exercising resize_token_embeddings()

* add docstring
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* implement gradient_accumulation_steps support in DeepSpeed integration

* typo

* cleanup

* cleanup

* free up memory at the end of train

* rework tests

* consistent formatting

* correction

* deepspeed integration

* style

* add test

* ds wants to do its own backward

* fp16 assert

* Update src/transformers/training_args.py
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* style

* for clarity extract what args are being passed to deepspeed

* introduce the concept of self.wrapped_model

* s/self.wrapped_model/self.model_wrapped/

* complete transition to self.wrapped_model / self.model

* fix

* doc

* give ds its own init

* add custom overrides, handle bs correctly

* fix test

* clean up model_init logic, fix small bug

* complete fix

* collapse --deepspeed_config into --deepspeed

* style

* start adding doc notes

* style

* implement hf2ds optimizer and scheduler configuration remapping

* oops

* call get_num_training_steps absolutely when needed

* workaround broken auto-formatter

* deepspeed_config arg is no longer needed - fixed in deepspeed master

* use hf's fp16 args in config

* clean

* start on the docs

* rebase cleanup

* finish up --fp16

* clarify the supported stages

* big refactor thanks to discovering deepspeed.init_distributed

* cleanup

* revert fp16 part

* add checkpoint-support

* more init ds into integrations

* extend docs

* cleanup

* unfix docs

* clean up old code

* imports

* move docs

* fix logic

* make it clear which file it's referring to

* document nodes/gpus

* style

* wrong format

* style

* deepspeed handles gradient clipping

* easier to read

* major doc rewrite

* Apply suggestions from code review
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* docs

* switch to AdamW optimizer

* style

* Apply suggestions from code review
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* clarify doc
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>