feat: better handle the multilabel classifiers by tweaking the response format

[m-misiura]

This PR improves the HuggingFace detector for multi-label sequence classification tasks:

1. When using AutoModelForSequenceClassification the detector now returns a ContentAnalysisResponse for each label whose score exceeds a configurable threshold, supporting multi-label outputs.
2. Users can exclude specific labels from triggering detections by specifying them in the non_trigger_labelsparameter.
3. The HuggingFace detector now imports scheme.py] from the common subdirectory to ensure schema consistency with other detectors.

Summary by Sourcery

Improve HuggingFace detector to support multi-label sequence classification by generating individual responses per label above a configurable threshold and excluding safe labels from detection through environment or request parameters

New Features:

• Emit separate ContentAnalysisResponse for each label exceeding threshold to support multi-label classification
• Enable exclusion of specified labels via SAFE_LABELS environment variable and per-request detector_params

Enhancements:

• Import common.scheme for schema consistency with other detectors
• Allow configurable detection threshold through detector_params

[sourcery-ai]

Reviewer's Guide

This PR refactors the HuggingFace detector to support threshold-based multi-label classification with environment- and parameter-based label exclusion, aligns schema imports for consistency, and exposes configurable detector parameters throughout.

Sequence diagram for multi-label sequence classification with threshold and label exclusion

sequenceDiagram
    participant User as actor User
    participant App as HuggingFace Detector
    User->>App: Submit ContentAnalysisHttpRequest (with text, detector_params)
    App->>App: process_sequence_classification(text, detector_params)
    App->>App: For each label, check if prob >= threshold and not in safe_labels
    App->>App: For each valid label, create ContentAnalysisResponse
    App-->>User: Return ContentsAnalysisResponse (multiple responses if multi-label)

Loading

Class diagram for updated Detector class with multi-label support

classDiagram
    class Detector {
        +risk_names: list
        +model: Any
        +cuda_device: Any
        +model_name: str
        +safe_labels: set
        +__init__()
        +process_causal_lm(text)
        +process_sequence_classification(text, detector_params=None, threshold=None)
        +run(input: ContentAnalysisHttpRequest) : ContentsAnalysisResponse
        +close()
    }
    class ContentAnalysisHttpRequest
    class ContentAnalysisResponse
    class ContentsAnalysisResponse
    Detector --> ContentAnalysisHttpRequest
    Detector --> ContentAnalysisResponse
    Detector --> ContentsAnalysisResponse

Loading

File-Level Changes

Change	Details	Files
Enable multi-label classification outputs	Changed process_sequence_classification signature to accept detector_params and threshold Replaced single-label prediction with loop to emit separate responses for each label above threshold Used dynamic detection_type derived from model.config.problem_type or label	detectors/huggingface/detector.py
Implement label exclusion via environment and parameters	Added _parse_safe_labels_env to read SAFE_LABELS from env with default Initialized self.safe_labels in init Merged env and detector_params safe_labels and filtered out matching labels in outputs	detectors/huggingface/detector.py
Refactor shared schema import	Switched import from local scheme to common.scheme for unified response types Removed obsolete detectors/huggingface/scheme.py file	detectors/huggingface/detector.py detectors/huggingface/scheme.py
Expose detector parameters in run flow	Updated run method to forward input.detector_params to process_sequence_classification	detectors/huggingface/detector.py

---

Tips and commands

Interacting with Sourcery

• Trigger a new review: Comment @sourcery-ai review on the pull request.
• Continue discussions: Reply directly to Sourcery's review comments.
• Generate a GitHub issue from a review comment: Ask Sourcery to create an
  issue from a review comment by replying to it. You can also reply to a
  review comment with @sourcery-ai issue to create an issue from it.
• Generate a pull request title: Write @sourcery-ai anywhere in the pull
  request title to generate a title at any time. You can also comment
  @sourcery-ai title on the pull request to (re-)generate the title at any time.
• Generate a pull request summary: Write @sourcery-ai summary anywhere in
  the pull request body to generate a PR summary at any time exactly where you
  want it. You can also comment @sourcery-ai summary on the pull request to
  (re-)generate the summary at any time.
• Generate reviewer's guide: Comment @sourcery-ai guide on the pull
  request to (re-)generate the reviewer's guide at any time.
• Resolve all Sourcery comments: Comment @sourcery-ai resolve on the
  pull request to resolve all Sourcery comments. Useful if you've already
  addressed all the comments and don't want to see them anymore.
• Dismiss all Sourcery reviews: Comment @sourcery-ai dismiss on the pull
  request to dismiss all existing Sourcery reviews. Especially useful if you
  want to start fresh with a new review - don't forget to comment
  @sourcery-ai review to trigger a new review!

Customizing Your Experience

Access your dashboard to:

• Enable or disable review features such as the Sourcery-generated pull request
  summary, the reviewer's guide, and others.
• Change the review language.
• Add, remove or edit custom review instructions.
• Adjust other review settings.

Getting Help

• Contact our support team for questions or feedback.
• Visit our documentation for detailed guides and information.
• Keep in touch with the Sourcery team by following us on X/Twitter, LinkedIn or GitHub.

[sourcery-ai]

Hey @m-misiura - I've reviewed your changes and they look great!

Prompt for AI Agents

Please address the comments from this code review:
## Individual Comments

### Comment 1
<location> `detectors/huggingface/detector.py:223` </location>
<code_context>
-        Returns:
-            List[ContentAnalysisResponse]: List of content analysis results.
-        """
+    def process_sequence_classification(self, text, detector_params=None, threshold=None):
+        detector_params = detector_params or {}
+        if threshold is None:
+            threshold = detector_params.get("threshold", 0.5)
+        # Merge safe_labels from env and request
+        request_safe_labels = set(detector_params.get("safe_labels", []))
+        all_safe_labels = set(self.safe_labels) | request_safe_labels 
         content_analyses = []
         tokenized = self.tokenizer(
</code_context>

<issue_to_address>
Combining safe_labels from environment and request may lead to type mismatches.

Because safe_labels may contain both ints and strings, and id2label keys are usually ints, mismatched types could cause incorrect filtering. Normalize types in all_safe_labels and comparison values to prevent subtle bugs.
</issue_to_address>

---

Sourcery is free for open source - if you like our reviews please consider sharing them ✨

• X
• Mastodon
• LinkedIn
• Facebook

_{Help me be more useful! Please click 👍 or 👎 on each comment and I'll use the feedback to improve your reviews.}

[sourcery-ai]

issue (bug_risk): Combining safe_labels from environment and request may lead to type mismatches.

Because safe_labels may contain both ints and strings, and id2label keys are usually ints, mismatched types could cause incorrect filtering. Normalize types in all_safe_labels and comparison values to prevent subtle bugs.