Repository navigation
AutoWeightLoader support Sglang native models 1: demo - #28671
Merged
Merged
Conversation
Contributor
|
Warning You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again! |
b8zhong
reviewed
Jul 7, 2026
| """Decorator to register an architecture-specific weight remap function. | ||
|
|
||
| The decorated function receives a model instance and returns a WeightsMapper | ||
| (or None if no remap is needed for this configuration). |
Collaborator
There was a problem hiding this comment.
Do we need it to return None? If no remap is needed, we'd better not use the decorator at all (to avoid confusion)
b8zhong
reviewed
Jul 7, 2026
| # --------------------------------------------------------------------------- | ||
|
|
||
|
|
||
| @dataclass |
Collaborator
There was a problem hiding this comment.
Can we use msgspec.Struct instead (codebase rule)
b8zhong
reviewed
Jul 7, 2026
b8zhong
left a comment
b8zhong
left a comment
Collaborator
There was a problem hiding this comment.
Can we show the use of WeightRemapRegistry in this PR (maybe modifying another model that requires this)
JD-ETH
commented
Jul 9, 2026
Document PR1–PR9 coverage plan, post_load protocol, unsupported models, and link from auto_loader module docstring. Co-authored-by: Cursor <cursoragent@cursor.com>
github-actions
Bot
added
the
documentation
Replace P2P/streaming framing with verify_complete asserting all expected runtime tensors were updated. Co-authored-by: Cursor <cursoragent@cursor.com>
- Use msgspec.Struct for StackedParamsDispatch; RemapRegistry requires WeightsMapper - Demo RemapRegistry on Llama v2 path with submodule stacked loaders - Parametrize stacked/remap unit tests with real Llama checkpoint names - Tighten v1/v2 equiv test: randomize init, compare all named_parameters - Drop SGLANG_ENABLE_WEIGHT_LOADER_V2 env comment Co-authored-by: Cursor <cursoragent@cursor.com>
Add registered SRTRunner tests for Qwen2 v1/v2 and transformers backend; restore manual get_model state_dict equiv test; drop test_auto_loader.py. Co-authored-by: Cursor <cursoragent@cursor.com>
Add wave-1 parallelism, sequencing for PR7–PR9, and infra conflict hotspots. Co-authored-by: Cursor <cursoragent@cursor.com>
Fridge003
reviewed
Jul 11, 2026
Fridge003
reviewed
Jul 11, 2026
This was referenced Jul 13, 2026
Remove design.md from the PR; point auto_loader docstring at the migration tracker issue linked from RFC sgl-project#24703. Co-authored-by: Cursor <cursoragent@cursor.com>
This was referenced Jul 15, 2026
Fridge003
approved these changes
Jul 16, 2026
Collaborator
|
/rerun-test test/manual/test_weight_loader_v2_equiv.py test/registered/model_loading/test_weight_loader_v2_e2e.py |
Contributor
|
Results for 🚀 ⛔ |
3 of 5 tasks
Jialin
added a commit
to Jialin/sglang
that referenced
this pull request
Jul 22, 2026
…gl-project#28671) sgl-project#28671 refactored LlamaForCausalLM/Qwen2ForCausalLM.load_weights to dispatch to self._legacy_load_weights / self._load_weights_v2. The reward and classification wrappers (llama_reward, llama_classification, qwen2_rm, qwen2_classification) borrow the base loader via an unbound call `Base.load_weights(self, ...)` but inherit only nn.Module, so `self` has no `_legacy_load_weights` and weight load fails with AttributeError (default path, since v2 is opt-in). Call `_legacy_load_weights` directly, restoring the pre-sgl-project#28671 behavior. These wrapped models are not AutoWeightLoader v2 citizens, so legacy is the correct path.
Contributor
Author
|
i will look into supporting the wrappers methods as well. |
Zhylkaaa
pushed a commit
to Zhylkaaa/sglang
that referenced
this pull request
Jul 29, 2026
Zhylkaaa
pushed a commit
to Zhylkaaa/sglang
that referenced
this pull request
Jul 29, 2026
…sgl-project#28671) (sgl-project#31988) Co-authored-by: Alex Nails <alex.nails@radixark.ai>
Atituiset
pushed a commit
to Atituiset/sglang
that referenced
this pull request
Sep 10, 2026
Atituiset
pushed a commit
to Atituiset/sglang
that referenced
this pull request
Sep 10, 2026
…sgl-project#28671) (sgl-project#31988) Co-authored-by: Alex Nails <alex.nails@radixark.ai>
3 of 5 tasks
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.
Motivation
AutoWeightLoader supports sglang models, better structural separation and guarantee.
Following discussion #24703 (comment)
Design and migration plan (PR1–PR9): #31051
Modifications
Speed Tests and Profiling
Checklist
CI States
Latest PR Test (Base): ❌ Run #29841391934
Latest PR Test (Extra): ❌ Run #29841391548