awslabs / awslabs/keys_values

Refactor node annotations in autograd_hooks.py

Open
#31 1 comment 0 reactions 0 assignees View on GitHub
enhancement
Dominant language
Python
Stars
18
Forks
8
PR merge metrics
No merged PRs in 30d

Description

**Is your feature request related to a problem? Please describe.**
Currently, node annotations are defined as a single dataclass, and creation and application are hacked into autograd_hooks.py and the training replay caches. This is not clean and hard to extend.

**Describe the solution you'd like**
Abstraction of node annotation as class, with `create`, `match`, and `apply` methods. Anything specific to a node annotation is provided there. All existing types are subclasses.

Contributor guide

Open the contributing guide

Research direction

Start by reading autograd_hooks.py and the training replay caches to locate the current node-annotation dataclass and its creation and application paths. Review the existing annotation types and determine how they can expose create, match, and apply methods as subclasses. Done means the existing annotation behavior is preserved while annotation-specific logic is moved out of those callers.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning
Issue type
Refactor
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
32/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.