[Feature] Add hooks in trainer #1249
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
HAOCHENYE
force-pushed
the
yehc/hooks
branch
2 times, most recently
from
72c02e4 to
df75250
Compare
HAOCHENYE
force-pushed
the
yehc/hooks
branch
from
df75250 to
039a00b
Compare
Copilot finished reviewing on behalf of
HAOCHENYE
November 21, 2025 07:31
Contributor
There was a problem hiding this comment.
Pull Request Overview
This PR adds a hooks system to the trainer, allowing users to inject custom logic at specific training stages. The hooks can be triggered after saving checkpoints (DCP, HF, snapshots) and after each training step.
Key changes:
- Introduced
HooksConfigwith four hook stages:after_save_dcp,after_save_hf,after_save_snapshot, andafter_train_step - Added Protocol definitions for type-safe hook implementations with optional
connect_trainermethod - Created
LossLogandOtherLogTypedDict classes to formalize logging data structures
Reviewed Changes
Copilot reviewed 7 out of 7 changed files in this pull request and generated 4 comments.
Show a summary per file
| File | Description |
|---|---|
| xtuner/v1/train/trainer.py | Added hook Protocol definitions, HooksConfig class, hook execution logic in fit and save methods, new properties for cur_epoch and total_epoch |
| xtuner/v1/engine/train_engine.py | Defined LossLog and OtherLog TypedDict classes, updated train_step return type annotation |
| xtuner/v1/engine/vision_compose_train_engine.py | Updated train_step return type annotation and added type annotations for loss_log and other_log |
| xtuner/v1/engine/init.py | Exported LossLog and OtherLog types |
| xtuner/v1/rl/base/worker.py | Added type ignore comments for type checker compatibility |
| xtuner/_testing/testcase.py | Added create_pg method to set LOCAL_RANK environment variable |
| tests/train/test_trainer.py | Added comprehensive test for hooks functionality |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
209
to
213
| def _validate_hooks( | ||
| cls, | ||
| value: list[HookProtocol] | HookProtocol | None, # noqa: E501 | ||
| ) -> list[Callable] | None: |
There was a problem hiding this comment.
Normal methods should have 'self', rather than 'cls', as their first parameter.
HAOCHENYE
force-pushed
the
yehc/hooks
branch
from
039a00b to
d3887ce
Compare
hhaAndroid
approved these changes
Nov 24, 2025
Collaborator
|
fix lint |
1. **`after_save_dcp`** - Triggered by `Trainer._maybe_save()` (non-snapshot saves) - Executes after saving distributed checkpoint 2. **`after_save_snapshot`** - Triggered by `Trainer._maybe_save()` when saving snapshot - Executes after creating training snapshot 3. **`after_save_hf`** - Triggered by `Trainer._maybe_save_hf()` - Executes after saving in HuggingFace format 4. **`after_train_step`** - Triggered after `TrainEngine.train_step()` completes - Executes after each training step - Hooks are optional and backward-compatible - Each hook receives context (step number, paths, etc.) - No performance impact when not registered
HAOCHENYE
force-pushed
the
yehc/hooks
branch
from
d3887ce to
cf18fa5
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.