Implementation:Norrrrrrr lyn WAInjectBench load jsonl

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

Overview

Concrete tool for loading JSONL training files into structured lists of features and labels, provided by the WAInjectBench training modules.

Description

The load_jsonl function exists in two variants:

• Text variant (train/embedding-t.py:L12-22): Parses {"text", "label", "source"} entries, returning (texts, labels, sources) tuples.
• Image variant (train/embedding-i.py:L16-25): Parses {"path", "label"} entries, returning (paths, labels) tuples.

Both variants skip empty lines and parse labels as integers.

Usage

Called at the beginning of classifier training to load labeled training data from JSONL files.

Code Reference

Source Location

• Repository: WAInjectBench
• File: train/embedding-t.py (L12-22), train/embedding-i.py (L16-25)

Signature

# Text variant (train/embedding-t.py:L12-22)
def load_jsonl(file_path):
    texts, labels, sources = [], [], []
    with open(file_path, "r", encoding="utf-8") as f:
        for line in f:
            if not line.strip():
                continue
            data = json.loads(line)
            texts.append(data["text"])
            labels.append(data["label"])
            sources.append(data.get("source", "unknown"))
    return texts, labels, sources

# Image variant (train/embedding-i.py:L16-25)
def load_jsonl(file_path):
    paths, labels = [], []
    with open(file_path, "r", encoding="utf-8") as f:
        for line in f:
            if not line.strip():
                continue
            data = json.loads(line)
            paths.append(data["path"])
            labels.append(int(data["label"]))
    return paths, labels

Import

# Defined locally in each training script
import json

I/O Contract

Inputs

Name	Type	Required	Description
file_path	str	Yes	Path to a JSONL training file

Outputs (Text Variant)

Name	Type	Description
texts	List[str]	Text content of each sample
labels	List[int]	Binary labels (0=benign, 1=malicious)
sources	List[str]	Source identifiers (default "unknown")

Outputs (Image Variant)

Name	Type	Description
paths	List[str]	Image file paths
labels	List[int]	Binary labels (0=benign, 1=malicious)

Usage Examples

Loading Text Training Data

# Text variant
texts, labels, sources = load_jsonl("train_data/text_dataset.jsonl")
print(f"Loaded {len(texts)} samples, {sum(labels)} malicious")

Loading Image Training Data

# Image variant
paths, labels = load_jsonl("train_data/image_dataset.jsonl")
print(f"Loaded {len(paths)} images, {sum(labels)} malicious")

Related Pages

Implements Principle

• Principle:Norrrrrrr_lyn_WAInjectBench_Training_Data_Loading