The model predicts the programming language of a source code snippet using a hybrid ensemble architecture combining a fine-tuned CodeBERT transformer with classical machine learning models.

Model Overview

The classifier combines multiple models to improve prediction accuracy:

CodeBERT Transformer
XGBoost
Random Forest
Gradient Boosting
Weighted Ensemble

The ensemble leverages transformer-based semantic understanding together with statistical TF-IDF features for robust language identification.

Supported Languages

The model is trained to classify:

Python
Q#
Qiskit
OpenQASM
Cirq

(The exact list depends on the training dataset.)

Architecture

Source Code
      │
      ▼
 ┌──────────────┐
 │  CodeBERT    │
 └──────────────┘
        │
        ▼

TF-IDF Features
        │
        ▼
 ┌──────────────┐
 │  XGBoost     │
 ├──────────────┤
 │ RandomForest │
 ├──────────────┤
 │GradientBoost │
 └──────────────┘
        │
        ▼
 Weighted Ensemble
        │
        ▼
 Predicted Language

Training

The model was trained using:

Hugging Face Transformers
Microsoft CodeBERT
Scikit-learn
XGBoost
PyTorch

Feature extraction:

TF-IDF
1–3 gram features
5,000 maximum features

Files

File	Description
codebert/	Fine-tuned CodeBERT model
tfidf.pkl	TF-IDF Vectorizer
xgboost.pkl	XGBoost classifier
random_forest.pkl	Random Forest classifier
gradient_boosting.pkl	Gradient Boosting classifier
label_encoder.pkl	Label encoder
ensemble_weights.json	Ensemble weights

Usage

python

from transformers import AutoTokenizer, AutoModelForSequenceClassification

tokenizer = AutoTokenizer.from_pretrained("yashodhajayasinghe/nexar-quantum-language-classifier")
model = AutoModelForSequenceClassification.from_pretrained("yashodhajayasinghe/nexar-quantum-language-classifier")

Intended Uses

Suitable for:

Programming language detection
Static code analysis
Source code indexing
Repository analytics
Intelligent developer tools
AI-assisted code understanding

Limitations

Performance may decrease for:

Extremely short code snippets
Mixed-language files
Obfuscated or minified code
Languages not included in the training dataset

Security Notice

This repository contains several serialized Scikit-learn models (.pkl files).

These files were generated using joblib and contain standard Scikit-learn model objects only.

Python pickle files are inherently executable during deserialization, therefore Hugging Face's automated malware scanner may display heuristic warnings. Users should only load pickle files from trusted sources.

Citation

If you use this model in your research or project, please cite:

Nexar Quantum Language Classifier
Nexar Quantum Code Analysis Engine
2026

License

This project is released under the MIT License.

Author

Yashodha Lasith Jayasinghe

Software Engineer | AI & Machine Learning Developer

GitHub: https://github.com/yashodalasith

Hugging Face: https://huggingface.co/yashodhajayasinghe