Commits · 3cbf754720856b255992e16e1c9e504d2734af2e · wuxk1 / Megatron-LM

"configs/_base_/models/mobilenet_v2_1x.py" did not exist on "8f9dd0edefa849b2552ba149141ddb369bdbec4e"

26 Feb, 2021 2 commits
- Fix deadlock when get_num_microbatches() < pipeline-parallel size (don't try... · 08ddde0c
  Deepak Narayanan authored Feb 26, 2021
```
Fix deadlock when get_num_microbatches() < pipeline-parallel size (don't try to measure pipeline stall)
```
  08ddde0c
- Support Torch DDP for single-stage, num_microbatches() > 1 · 65eeb427
  Deepak Narayanan authored Feb 24, 2021
  
  65eeb427
13 Feb, 2021 1 commit
- More comments and some cleanup (e.g., better variable names) · 5489bda9
  Deepak Narayanan authored Feb 13, 2021
  
  5489bda9
10 Feb, 2021 2 commits
- Comments in megatron/schedules.py and address a few more comments · 626645c0
  Deepak Narayanan authored Feb 10, 2021
  
  626645c0
- Move unwrap to megatron/utils.py and clean up imports in megatron/schedules.py · cc691cbf
  Deepak Narayanan authored Feb 10, 2021
  
  cc691cbf
09 Feb, 2021 2 commits

Put in barriers in appropriate places to measure length of pipeline stall · 8e922d5b
Deepak Narayanan authored Jan 09, 2021

8e922d5b

Interleaved pipeline execution and code refactoring · dd889062

Deepak Narayanan authored Dec 12, 2020

- 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

dd889062