FSDP2 tutorial#3358
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/tutorials/3358
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 New FailureAs of commit 9e7d160 with merge base 78933b1 ( NEW FAILURE - The following job has failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
FSDP2 tutorial is ready for review @AlannaBurke @svekars |
|
link failure is expected for "https://docs.pytorch.org/tutorials/intermediate/FSDP1_tutorial.html". It will work when we land FSDP1_tutorial in this PR |
|
|
||
| # initialize the process group | ||
| dist.init_process_group("nccl", rank=rank, world_size=world_size) | ||
| ``fully_shard`` register forward/backward hooks to all-gather parameters before computation, and reshard parameters after computation. To overlap all-gathers with computation, FSDP2 offers **implicit prefetching** that works out of the box with the training loop above and **explicit prefetching** for advanced users to control all-gather schedules manually. |
There was a problem hiding this comment.
register -> registers
reshard -> reshards
| **Author**: `Hamid Shojanazeri <https://github.com/HamidShojanazeri>`__, `Yanli Zhao <https://github.com/zhaojuanmao>`__, `Shen Li <https://mrshenli.github.io/>`__ | ||
|
|
||
| .. note:: | ||
| |edit| FSDP1 is deprecated. Please check out `FSDP2 tutorial <https://docs.pytorch.org/tutorials/intermediate/FSDP_tutorial.html>`_. |
There was a problem hiding this comment.
| |edit| FSDP1 is deprecated. Please check out `FSDP2 tutorial <https://docs.pytorch.org/tutorials/intermediate/FSDP_tutorial.html>`_. | |
| FSDP1 is deprecated. Please check out `FSDP2 tutorial <https://docs.pytorch.org/tutorials/intermediate/FSDP_tutorial.html>`_. |
|
@weifengpy Can you post the preview link for this tutorial? |
I got the preview link here https://docs-preview.pytorch.org/pytorch/tutorials/3358/intermediate/FSDP_tutorial.html |
5 participants
FSDP2 tutorial replaces FSDP1 tutorial in place (intermediate_source/FSDP_tutorial.rst)
FSDP1 tutorial is renamed to intermediate_source/FSDP1_tutorial.rst. FSDP2 tutorial link to it
the code for this tutorial is commited to pytorch examples, https://github.com/pytorch/examples/tree/main/distributed/FSDP2