[Feature] mm and thinking model support structred output #2749
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
|
Thanks for your contribution! |
kevincheng2
changed the title
kevincheng2
force-pushed
the
mm_structred_output
branch
from
d07f737 to
72de4a3
Compare
Contributor
There was a problem hiding this comment.
Pull Request Overview
This PR adds structured output support via guided decoding (reasoning parsers) for multi-modal and thinking models, including offline inference capabilities.
- Introduce a new
--reasoning_parserCLI argument and propagate it through configuration to model runners. - Extend the sampling and guided decoding pipeline: updated
Sampler, guided backend interfaces, and skip-index logic. - Enhance
SamplingParamswithGuidedDecodingParamsand document offline inference usage for structured outputs.
Reviewed Changes
Copilot reviewed 18 out of 18 changed files in this pull request and generated 2 comments.
Show a summary per file
| File | Description |
|---|---|
| fastdeploy/worker/worker_process.py | Add --reasoning_parser CLI arg and integrate it into FDConfig. |
| fastdeploy/worker/vl_gpu_model_runner.py | Initialize guided backend and reasoning parser; update guided decoding flow in the GPU model runner. |
| fastdeploy/model_executor/layers/sample/sampler.py | Enhance Sampler to support reasoning parsing and skip indices when masking tokens. |
| fastdeploy/engine/sampling_params.py | Introduce GuidedDecodingParams in SamplingParams for offline structured inference. |
| docs/features/structured_outputs.md | Add offline inference examples for structured output using GuidedDecodingParams. |
Comments suppressed due to low confidence (3)
fastdeploy/worker/vl_gpu_model_runner.py:145
- The code checks for
guided_json,guided_regex,guided_grammar, andstructural_tagbut does not handleguided_choicefromGuidedDecodingParams. Add support forguided_choiceto ensure all constraint types are honored.
elif request.guided_grammar is not None:
fastdeploy/engine/engine.py:1049
- The code references
self.cfg.reasoning_parser, butreasoning_parseris not defined on the engine config object. It should likely referenceself.cfg.model_config.reasoning_parser.
f" --reasoning_parser {self.cfg.reasoning_parser}")
fastdeploy/worker/vl_gpu_model_runner.py:152
- Using
request.get(...)may not work ifrequestis not a dict-like object. Consider usinggetattr(request, 'enable_thinking', True)to access the attribute safely.
enable_thinking=request.get("enable_thinking", True),
aac8503 to
04c2f3c
Compare
Jiang-Jia-Jun
previously approved these changes
Jul 18, 2025
kevincheng2
force-pushed
the
mm_structred_output
branch
3 times, most recently
from
2ef373a to
69fc3a2
Compare
kevincheng2
force-pushed
the
mm_structred_output
branch
from
69fc3a2 to
6bd3676
Compare
kevincheng2
force-pushed
the
mm_structred_output
branch
from
0429910 to
3e9bba5
Compare
kevincheng2
force-pushed
the
mm_structred_output
branch
from
aec275d to
278d3bd
Compare
kevincheng2
force-pushed
the
mm_structred_output
branch
from
8f9fb63 to
9ba1d41
Compare
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## develop #2749 +/- ##
==========================================
Coverage ? 39.02%
==========================================
Files ? 11
Lines ? 123
Branches ? 19
==========================================
Hits ? 48
Misses ? 69
Partials ? 6
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
Jiang-Jia-Jun
approved these changes
Sep 2, 2025
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.
Uh oh!
There was an error while loading. Please reload this page.