Accelerate scaler #2677

avishniakov · 2024-05-07T13:26:53Z

Describe changes

This PR introduces the concept of Scalers, which can be passed into the steps' definitions to allow step run on some parallelization engine. Scalers included here:

AccelerateScaler (a major goal of this PR): allows to run a step function via accelerate run... without any effort for rewriting the step (except the fact, that it should properly handle accelerate by itself, like save in the main process, etc.)
AggregateScaler - this is more of a demo of what we can do for other use cases. This scaler just parallelizes the step logic locally and later aggregates back the results using the chosen aggregate function.

Now I would like to gather some early feedback from you.

Examples

from zenml import step
from zenml.integrations.accelerate import AccelerateScaler

@step(scaler=AccelerateScaler(num_processes=42))
def training_step(some_param: int, ...):
    # your training code is below
    ...

from zenml import step, pipeline
from zenml.scalers import AggregateScaler

@step(scaler=AggregateScaler(parameters={"a":[1,2,3],"b":[4,5,6]}, agg_function="sum"))
def training_step_with_sum_aggregation(a:int = None, b:int = None, c:int = 2)->int:
    # your code is below
    return a+b+c

@pipeline
def pipeline_with_aggregate_scaler():
    training_step_with_sum_aggregation(c=3)
# actual step output would be (1+4+3)+(2+5+3)+(3+6+3) = 30,
# where last "+3" comes from constant `c` parameter

To be done still in this PR:

Tests
Docs
Make accelerate a proper integration

How this works in the wild: zenml-io/zenml-projects#102

Pre-requisites

Please ensure you have done the following:

I have read the CONTRIBUTING.md document.
If my change requires a change to docs, I have updated the documentation accordingly.
I have added tests to cover my changes.
I have based my new branch on develop and the open PR is targeting develop. If your branch wasn't based on develop read Contribution guide on rebasing branch to develop.
If my changes require changes to the dashboard, these changes are communicated/requested.

Types of changes

Bug fix (non-breaking change which fixes an issue)
New feature (non-breaking change which adds functionality)
Breaking change (fix or feature that would cause existing functionality to change)
Other (add details above)

coderabbitai · 2024-05-07T13:26:57Z

Important

Auto Review Skipped

Auto reviews are disabled on this repository.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Thank you for using CodeRabbit. We offer it for free to the OSS community and would appreciate your support in helping us grow. If you find it useful, would you consider giving us a shout-out on your favorite social media?

Share

Tips

Chat

There are 3 ways to chat with CodeRabbit:

Review comments: Directly reply to a review comment made by CodeRabbit. Example:
- I pushed a fix in commit <commit_id>.
- Generate unit testing code for this file.
- Open a follow-up GitHub issue for this discussion.
Files and specific lines of code (under the "Files changed" tab): Tag @coderabbitai in a new review comment at the desired location with your query. Examples:
- @coderabbitai generate unit testing code for this file.
- @coderabbitai modularize this function.
PR comments: Tag @coderabbitai in a new PR comment to ask questions about the PR branch. For the best results, please provide a very specific query, as very limited context is provided in this mode. Examples:
- @coderabbitai generate interesting stats about this repository and render them as a table.
- @coderabbitai show all the console.log statements in this repository.
- @coderabbitai read src/utils.ts and generate unit testing code.
- @coderabbitai read the files in the src/scheduler package and generate a class diagram using mermaid and a README in the markdown format.

Note: Be mindful of the bot's finite context window. It's strongly recommended to break down tasks such as reading entire modules into smaller chunks. For a focused discussion, use review comments to chat about specific files and their changes, instead of using the PR comments.

CodeRabbit Commands (invoked as PR comments)

@coderabbitai pause to pause the reviews on a PR.
@coderabbitai resume to resume the paused reviews.
@coderabbitai review to trigger a review. This is useful when automatic reviews are disabled for the repository.
@coderabbitai resolve resolve all the CodeRabbit review comments.
@coderabbitai help to get help.

Additionally, you can add @coderabbitai ignore anywhere in the PR description to prevent this PR from being reviewed.

CodeRabbit Configration File (`.coderabbit.yaml`)

You can programmatically configure CodeRabbit by adding a .coderabbit.yaml file to the root of your repository.
Please see the configuration documentation for more information.
If your editor has YAML language server enabled, you can add the path at the top of this file to enable auto-completion and validation: # yaml-language-server: $schema=https://coderabbit.ai/integrations/schema.v2.json

Documentation and Community

Visit our Documentation for detailed information on how to use CodeRabbit.
Join our Discord Community to get help, request features, and share feedback.
Follow us on X/Twitter for updates and announcements.

avishniakov · 2024-05-14T06:58:52Z

@htahir1 not putting you to reviewers, but you might have what to add to this story.

strickvl · 2024-05-14T08:35:09Z

src/zenml/integrations/accelerate/__init__.py

is this an accelerate integration or is this just part of our huggingface integration? I'm pretty sure you'll get accelerate already with the packages we have defined there?

strickvl · 2024-05-14T08:38:52Z

src/zenml/utils/function_utils.py

+    return False
+
+
+def _is_valid_optional_arg(arg_type: Any) -> bool:


Suggested change

def _is_valid_optional_arg(arg_type: Any) -> bool:

def _is_valid_optional_arg(arg_type: Any) -> bool:

"""Check if the given argument type is a valid Optional type.

A valid Optional type is defined as a Union with two arguments, where:

- The first argument is either an allowed type or a valid collection type.

- The second argument is the NoneType.

Args:

arg_type: The type to check.

Returns:

True if the argument type is a valid Optional type, False otherwise.

"""

htahir1 · 2024-05-14T09:49:53Z

Wow what a great idea. I havnt look to deep but what is the difference between this and step operators?

avishniakov · 2024-05-14T10:22:38Z

Wow what a great idea. I havnt look to deep but what is the difference between this and step operators?

Thx, IMO, going forward one can define a step which, for example, takes some slicing parameters and inside the step, some intense data processing is happening and instead of running it with step operator k8s or other, I say scale it using step operator k8s or other.
This is not yet implemented for sure or shaped well, but this is how I see this: a mixture of specific libs and the reuse of some step operators to give a performance boost in parallelisable operations.
Does that answer?

htahir1 · 2024-05-14T13:09:48Z

The only reservation I have is that there are simply too many concepts in ZenML and this is a new one that I haven't really understood yet

avishniakov · 2024-05-14T13:16:25Z

The only reservation I have is that there are simply too many concepts in ZenML and this is a new one that I haven't really understood yet

That's very valid, based on the feedback we saw. Do you feel that saying AccelerateStepOperator is a better option here? On the other hand, it shall be still backed by VertexStepOperator or another GPU compute, which makes things really complicated to implement.

avishniakov · 2024-06-13T12:36:15Z

Closed in favor of #2746

avishniakov added 9 commits May 6, 2024 15:59

sugar for accelerate

004b77c

fix cli command format

95fa2c0

properly process collections

b5f0279

properly process collections

ae012de

properly process collections

10135e2

properly process collections

ecd4170

properly process collections

c80d348

properly process collections

621f7f8

lint

d7388c6

github-actions bot added internal To filter out internal PRs and issues enhancement New feature or request labels May 7, 2024

avishniakov mentioned this pull request May 7, 2024

Multi GPU with PEFT on LLM zenml-io/zenml-projects#102

Merged

avishniakov added 14 commits May 8, 2024 10:57

add scalers setting to the step

39dd1ac

properly store ref to original scaler

c250e3a

properly store ref to original scaler

ce2386e

clii wrapper as contextmanager

8f1d2a9

cli wrapper as contextmanager

84a7e49

cli wrapper as contextmanager

4ecff7d

return from accelerate command

c01597d

cli wrapper as contextmanager

6a43759

cli wrapper as contextmanager

86045b1

proper paths

e4ed1e2

add debug outputs

88bfe64

add debug outputs

963c819

return from CLICK

5f06ddd

add aggregate scaler

52a81f9

avishniakov requested review from schustmi and strickvl May 14, 2024 06:58

avishniakov changed the title ~~[WIP] Accelerate jobs automation~~ Accelerate scaler May 14, 2024

strickvl reviewed May 14, 2024

View reviewed changes

avishniakov closed this Jun 13, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Accelerate scaler #2677

Accelerate scaler #2677

avishniakov commented May 7, 2024 •

edited

Loading

coderabbitai bot commented May 7, 2024

Auto Review Skipped

Chat

CodeRabbit Commands (invoked as PR comments)

CodeRabbit Configration File (`.coderabbit.yaml`)

Documentation and Community

avishniakov commented May 14, 2024

strickvl May 14, 2024

strickvl May 14, 2024

htahir1 commented May 14, 2024

avishniakov commented May 14, 2024

htahir1 commented May 14, 2024

avishniakov commented May 14, 2024

avishniakov commented Jun 13, 2024

		return False


		def _is_valid_optional_arg(arg_type: Any) -> bool:

-def _is_valid_optional_arg(arg_type: Any) -> bool:
+def _is_valid_optional_arg(arg_type: Any) -> bool:
+    """Check if the given argument type is a valid Optional type.
+    A valid Optional type is defined as a Union with two arguments, where:
+    - The first argument is either an allowed type or a valid collection type.
+    - The second argument is the NoneType.
+    Args:
+        arg_type: The type to check.
+    Returns:
+        True if the argument type is a valid Optional type, False otherwise.
+    """

Accelerate scaler #2677

Accelerate scaler #2677

Conversation

avishniakov commented May 7, 2024 • edited Loading

Describe changes

Examples

To be done still in this PR:

Pre-requisites

Types of changes

coderabbitai bot commented May 7, 2024

Auto Review Skipped

Chat

CodeRabbit Commands (invoked as PR comments)

CodeRabbit Configration File (.coderabbit.yaml)

Documentation and Community

avishniakov commented May 14, 2024

strickvl May 14, 2024

Choose a reason for hiding this comment

strickvl May 14, 2024

Choose a reason for hiding this comment

htahir1 commented May 14, 2024

avishniakov commented May 14, 2024

htahir1 commented May 14, 2024

avishniakov commented May 14, 2024

avishniakov commented Jun 13, 2024

avishniakov commented May 7, 2024 •

edited

Loading

CodeRabbit Configration File (`.coderabbit.yaml`)