argonne-lcf / Megatron-DeepSpeed

Ongoing research training transformer language models at scale, including: BERT & GPT-2
Other
9 stars 12 forks source link

Create `alcf-startup-time` #34

Closed saforem2 closed 4 months ago

saforem2 commented 4 months ago

Profiling deepspeed.initialize(...)

Original Plan: ~~Wraps initialization in a~~ ```python import ezpz as ez RANK = ez.get_rank() cm = ez.profile.get_context_manager(rank=RANK) with cm: # Megatron-DeepSpeed init, create model, data loaders, etc. ``` ~~see:~~ - ~~[`saforem2/ezpz/src/ezpz/profile.py`](https://github.com/saforem2/ezpz/blob/main/src/ezpz/profile.py)~~ - ~~[\[PR\]: Merge ezpz-profile into main #12](https://github.com/saforem2/ezpz/pull/12#issue-2366943397)~~ ~~for additional implementation details~~

Instead (see discussion):

[^ezpz]: From ezpz.profile