fix(output_parsers): iterable unpacking compatibility in content safety parsers #1242
+708
−27
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
There was a problem hiding this comment.
Pull Request Overview
This PR fixes iterable unpacking compatibility issues in output parsers by switching the return type from tuples to list-based sequences. Key changes include:
- Updating output parser functions (is_content_safe, nemoguard_parse_prompt_safety, nemoguard_parse_response_safety) to return Sequence[Union[bool, str]].
- Adding a helper function (_parse_unsafe_violations) to robustly parse policy violations.
- Updating integration and unit tests to verify proper unpacking and correct behavior in edge-case scenarios.
Reviewed Changes
Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.
| File | Description |
|---|---|
| tests/test_output_parsers.py | Removed obsolete user intent test to focus solely on content safety. |
| tests/test_content_safety_output_parsers.py | Added comprehensive tests for both JSON and plain-text responses covering multiple safety cases. |
| tests/test_content_safety_integration.py | Added integration tests ensuring compatibility of the new list-based parser outputs with starred unpacking. |
| nemoguardrails/llm/output_parsers.py | Updated parser functions to return lists and introduced a helper to maintain case-preservation for violations. |
Comments suppressed due to low confidence (1)
nemoguardrails/llm/output_parsers.py:149
- When 'Safety Categories' is present but is an empty string, consider returning a list with an empty string (i.e. ['']) instead of an empty list to align with the expected behavior in tests.
}
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## develop #1242 +/- ##
===========================================
+ Coverage 69.04% 69.51% +0.47%
===========================================
Files 161 161
Lines 15995 16012 +17
===========================================
+ Hits 11043 11131 +88
+ Misses 4952 4881 -71
Flags with carried forward coverage won't be shown. Click here to find out more.
🚀 New features to boost your workflow:
|
858561a to
435c772
Compare
|
would you please review the test cases to see if the expected behavior is OK? The interface of output parser has changed and it requires us to use iterable unpacking is_safe, *violated_policies = is_content_safe(response)The safe/unsafe is always the first element of the sequence. if a response is considered safe, violations will always be an empty list. Is this assumption correct? if not, do we need to modify the safe/unsafe logic? (see https://github.com/NVIDIA/NeMo-Guardrails/pull/1242/files#r2169265190) |
Pouyanpi
commented
Jun 26, 2025
| context=context, | ||
| ) | ||
|
|
||
| # TODO: @trebedea @prasoonvarshney is this the expected behavior? |
There was a problem hiding this comment.
…ent safety parsers - Change return type from Tuple[bool, List[str]] to Sequence[Union[bool, str]] - Fix empty violations bug that returned [""] instead of [] - Preserve original case in violation parsing (S1 vs s1) - Add helper function _parse_unsafe_violations() for robust parsing - Remove duplicate test file tests/test_output_parsers.py - Add comprehensive test coverage (38 tests) in test_content_safety_output_parsers.py Fixes starred unpacking syntax: is_safe, *violated_policies = result Previously failed due to tuple format incompatibility with actions expecting list format. Closes #1241 refactor tests
d60a06e to
a3b1cb6
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Problem
Content safety actions were failing due to output parser interface incompatibility. The parsers returned
Tuple[bool, List[str]]format, but actions used starred unpacking syntaxis_safe, *violated_policies = resultwhich expected list format.Root Cause
Multiple bugs in output parsers:
(True, [])incompatible withis_safe, *violated_policies = resultthat is introduced in fix(content_safety): replace try-except with iterable unpacking for policy violations #1207[""]instead of[]for no violations"s1"vs"S1")Solution
Tuple[bool, List[str]]→Sequence[Union[bool, str]][True]or[False, "policy1", "policy2"]format"S1"stays"S1", not"s1"_parse_unsafe_violations()for robust parsingImpact
Breaking Changes
None: this is a bug fix that makes the newly introduced interable unpacking syntax work as intended.
Closes #1241