GitLab

- Split a model's computation into multiple virtual stages as needed,
and schedule communication correctly between these virtual stages
- Move schedule code into `schedules.py` and communication code into
`p2p_communication.py`
- Use hyphens instead of spaces in all time logging for consistency
- Factor out code in megatron/training.py into helper functions
- Refactor evaluate() function: make it use forward_backward_schedule
functions

Refactor code according to Jared's comments: move pipelining and non-pipelining training loops into separate methods

Also, use mpu.get_*_model_parallel_size() instead of args.*_model_parallel_size

Also includes following changes for inter-layer model-parallel implementation:
- Refactoring of model implementations
- Training loop changes to support inter-layer communication using `ring_exchange`
- New groups for inter-layer communication
- Checkpoint changes
- Command line arguments

Note that there is currently an issue with bert using Torch DDP.

Co-authored-by: shoeybi <shoeybim@gmail.com>
Co-authored-by: raulpuric <raulpuric@berkeley.edu>
Co-authored-by: jaredcasper <jaredcasper@gmail.com>
Co-authored-by: mpatwary <mostofa.patwary@gmail.com>
Co-authored-by: plegresl <plegresl@gmail.com>

* threaded tf_dl+presplit sentences+shuffled dataset with resume

* elaborate in readme