Skip to content

Commit

Permalink
[Doc] round_robin_gradients (#1261)
Browse files Browse the repository at this point in the history
* Fix docstring

* Make screenshots clickable for easier viewing

* Navigation menu in alphabetical order; More clicable screenshots

* Rename 1Cycle doc

* Tweak naming

* Remove no longer used flag

* ZeRO3 Offload release

* Single GPU results

* Rearrange figures

* Single GPU text

* tweak intro

* zero3-offload section

* Add asynchronous i/o docs

* Fix print_per_steps doc

* Document round_robin_gradients

* Tweak description

* Trigger CI
  • Loading branch information
tjruwase authored Jul 30, 2021
1 parent e82060d commit 40c381d
Showing 1 changed file with 7 additions and 0 deletions.
7 changes: 7 additions & 0 deletions docs/_pages/config-json.md
Original file line number Diff line number Diff line change
Expand Up @@ -301,6 +301,7 @@ Enabling and configuring ZeRO memory optimizations
"elastic_checkpoint" : [true|false],
"stage3_gather_fp16_weights_on_model_save": [true|false],
"ignore_unused_parameters": [true|false]
"round_robin_gradients": [true|false]
}
```

Expand Down Expand Up @@ -358,6 +359,12 @@ Enabling and configuring ZeRO memory optimizations
| ------------------------------------------------------------------------------------------------------------------------------------------ | ------- |
| For use with ZeRO stage 1, enable backward hooks to reduce gradients during the backward pass or wait until the end of the backward pass. | `True` |

***round_robin_gradients***: [boolean]

| Description | Default |
| ------------------------------------------------------------------------------------------------------------------------------------------ | ------- |
| Stage 2 optimization for CPU offloading that parallelizes gradient copying to CPU memory among ranks by fine-grained gradient partitioning. Performance benefit grows with gradient accumulation steps (more copying between optimizer steps) or GPU count (increased parallelism). | `False` |

***offload_param***: [dictionary]

| Description | Default |
Expand Down

0 comments on commit 40c381d

Please sign in to comment.