Implementation:Norrrrrrr lyn WAInjectBench joblib dump

Knowledge Sources	WAInjectBench Joblib
Domains	Machine_Learning, Model_Management
Last Updated	2026-02-14 16:00 GMT

Overview

Concrete tool for serializing trained LogisticRegression classifiers to .pkl files, provided by the joblib library as used in the WAInjectBench embedding trainers.

Description

Both text and image embedding trainers use joblib.dump(clf, save_path) to persist the fitted classifier. The save path is constructed as {output_dir}/{jsonl_stem}_logreg.pkl. The text variant also optionally saves embeddings to a separate JSONL file (save_emb=True by default).

Usage

Called as the final step of train_single_classifier after fitting the LogisticRegression model.

Code Reference

Source Location

• Repository: WAInjectBench
• File: train/embedding-t.py (L38), train/embedding-i.py (L59-60)

Signature

# Text variant (train/embedding-t.py:L36-38)
model_name = Path(jsonl_file).stem + "_logreg.pkl"
save_path = os.path.join(output_dir, model_name)
joblib.dump(clf, save_path)

# Image variant (train/embedding-i.py:L58-60)
model_name = Path(jsonl_file).stem + "_logreg.pkl"
save_path = os.path.join(output_dir, model_name)
joblib.dump(clf, save_path)

Import

import joblib

I/O Contract

Inputs

Name	Type	Required	Description
clf	LogisticRegression	Yes	Fitted classifier object
save_path	str	Yes	Output file path ({output_dir}/{stem}_logreg.pkl)

Outputs

Name	Type	Description
.pkl file	File	Serialized classifier at the specified path

Usage Examples

Saving and Loading a Classifier

import joblib
from sklearn.linear_model import LogisticRegression

# Train classifier
clf = LogisticRegression(max_iter=1000)
clf.fit(embeddings, labels)

# Save
joblib.dump(clf, "models/dataset_logreg.pkl")

# Load for inference
loaded_clf = joblib.load("models/dataset_logreg.pkl")
predictions = loaded_clf.predict(new_embeddings)

Related Pages

Implements Principle

• Principle:Norrrrrrr_lyn_WAInjectBench_Model_Serialization