Skip to content

fix: route text analysis through production trained classifier #24

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
hrx01-dev merged 1 commit into from
Aug 27, 2026
Merged

fix: route text analysis through production trained classifier #24

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
15 changes: 4 additions & 11 deletions backend/production_main.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,8 +19,6 @@
# ---------------------------------------------------------------------------
# Canonical trained text classifier
# ---------------------------------------------------------------------------
# Keep the default model small enough for CPU deployments. The model is a
# genuine fine-tuned fake-news classifier, not a zero-shot classifier.
TEXT_MODEL_ID = os.getenv(
"TEXT_MODEL_ID",
"mrm8488/bert-tiny-finetuned-fake-news-detection",
Expand All @@ -37,7 +35,6 @@ def _load_production_text_model():
if secure.pipeline is None:
raise RuntimeError("transformers is not installed")

# top_k=None returns both class scores instead of only the winning class.
_TEXT_MODEL = secure.pipeline(
"text-classification",
model=TEXT_MODEL_ID,
Expand All @@ -55,8 +52,6 @@ def _text_scores(text: str) -> Dict[str, float]:
classifier = _load_production_text_model()
raw = classifier(text.strip())

# Transformers can return either [{...}, {...}] or [[{...}, {...}]]
# depending on the installed pipeline version and batching behaviour.
if raw and isinstance(raw[0], list):
raw = raw[0]

Expand All @@ -71,9 +66,6 @@ def _text_scores(text: str) -> Dict[str, float]:
elif "real" in label or "true" in label or label in {"label_1", "1"}:
real = score

# The selected model is a binary fake-news classifier. Some Transformers
# versions expose generic LABEL_0/LABEL_1 names, so retain the established
# model mapping as a fallback when semantic labels are unavailable.
if fake is None and real is None and raw:
first = raw[0]
label = str(first.get("label", "")).strip().lower()
Expand Down Expand Up @@ -121,11 +113,12 @@ def _production_content_score(text: str) -> Dict[str, float]:
return _text_scores(text)


# Replace the legacy zero-shot path and the secure wrapper's older adapter.
# Both /api/analyze/text and /api/analyze/url resolve these functions at
# request time, so they now use exactly the same trained classifier.
# The secure wrapper defines /api/analyze/text itself and resolves its model
# loader from the secure_main module. Patch that loader too; patching only the
# legacy module leaves the direct text route on the old model implementation.
legacy._load_text_classifier = _production_text_classifier
legacy._model_content_score = _production_content_score
secure._load_text_classifier = _production_text_classifier
secure._TEXT_MODEL_ID = TEXT_MODEL_ID
secure._TEXT_MODEL = None

Expand Down
Loading