feat(train): Add SequenceLength support for SFT, DPO, RLVR, RLAIF trainers#5898
Open
guanweim wants to merge 10 commits into
Open
feat(train): Add SequenceLength support for SFT, DPO, RLVR, RLAIF trainers#5898guanweim wants to merge 10 commits into
guanweim wants to merge 10 commits into
Conversation
…iners Add optional sequence_length parameter to all four trainers that enables customers to specify their desired context length for serverless training jobs. The parameter is passed in ServerlessJobConfig for recipe filtering. During trainer initialization, _get_fine_tuning_options_and_model_arn filters recipes by SequenceLength field, picking the smallest recipe with context length >= the requested value. Raises ValueError if no sufficient recipe exists or if recipes lack SequenceLength metadata. Changes: - ServerlessJobConfig: add sequence_length field - _parse_context_length: parse values like '8K' to integers - _get_fine_tuning_options_and_model_arn: filter by SequenceLength - _create_serverless_config: conditionally include sequence_length - SFTTrainer, DPOTrainer, RLVRTrainer, RLAIFTrainer: accept and thread sequence_length through init and train methods - Unit tests for all new functionality
… edit Add SequenceLength to service-2.json and regenerate shapes.py via codegen (python -m sagemaker.core.tools.codegen) instead of editing shapes.py manually.
…h not provided Keep the existing `next(...)` logic untouched for the default case (no sequence_length). Only build the candidates list and filter when sequence_length is explicitly requested, ensuring zero behavioral change for existing callers.
zachgk
reviewed
May 28, 2026
| elif (isinstance(training_type, TrainingType) and training_type == TrainingType.FULL) or training_type == "FULL": | ||
| recipe = next((r for r in recipes_with_template if not r.get("Peft") and not r.get("IsSubscriptionModel")), None) | ||
|
|
||
| # Override recipe selection when sequence_length is explicitly requested |
Contributor
There was a problem hiding this comment.
This is cleaner to think about as a recipe filter. So move the sequence_length filter above the recipe selection logic and reduce the list of available recipes. Afterwards the existing logic can stay the same
| accept_eula=accept_eula, | ||
| ) | ||
| if sequence_length is not None: | ||
| config_kwargs["sequence_length"] = sequence_length |
Contributor
There was a problem hiding this comment.
You can probably pass on the None value to ServerlessJobConfig
Contributor
mollyheamazon
left a comment
mollyheamazon
left a comment
There was a problem hiding this comment.
Please add integ test as sanity check as this is working e2e.
| training_type, | ||
| sagemaker_session, | ||
| sequence_length=None, | ||
| hub_name: str = "SageMakerPublicHub" |
Contributor
There was a problem hiding this comment.
This should be set to hub_name: Optional[str] = None for original pattern and consistency.
…config - Move sequence_length filtering above recipe selection to reduce recipes_with_template before existing logic runs - Always pass sequence_length to ServerlessJobConfig (no None guard)
Use Optional[str] = None instead of hardcoded "SageMakerPublicHub" default, letting get_sagemaker_hub_name() resolve it at runtime.
- _parse_context_length with invalid K value and non-numeric string - sequence_length filter when no recipes have SequenceLength field - sequence_length filter with FULL training type - verify smallest sufficient sequence_length is selected - verify no sequence_length uses first recipe (backward compat)
Author
added |
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 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.
Description of changes:
Summary
sequence_lengthparameter to SFTTrainer, DPOTrainer, RLVRTrainer, and RLAIFTrainer for specifying context length in serverless training jobsSequenceLengthfield, picking the smallest recipe with context length >= the requested valueSequenceLengthtoservice-2.jsonand regenerate shapes via codegenTest plan
_parse_context_lengthhelper (K suffix, lowercase, integer, None, empty)sequence_lengththrough_create_serverless_configwith/without sequence_lengthBy submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.