---
language: en
license: mit
tags:
- text-classification
- pytorch
---

# Comments Classifier (RuBERT fine-tune)

A Russian-language comment classification model fine-tuned on top of **RuBERT**. Developed as part of the Lubarsky Comments Model project.

## Overview

The model was fine-tuned on a labeled dataset of Russian-language comments. Its goal is to automatically determine the category/type of a given comment.

The repository contains three ready-to-use **standalone applications** built with **PyInstaller** — no Python installation or dependencies required:

| File | Size | Description |
|---|---|---|
| `run_trainer.zip` | ~2.6 GB | Application for fine-tuning the model |
| `run_prediction.zip` | ~2.5 GB | Application for running predictions |
| `run_classifier.zip` | ~60 MB | Application for manual comment classification |
| `QA_dataset.csv` | ~75 kB | Quality assurance dataset |

---

## Quick Start

> ⚠️ **No Python installation required** — all three programs are self-contained `.exe` applications.

### 1. Download the ZIP archive

Download one or more archives from this page.

### 2. Extract the archive

Extract the downloaded archive to a convenient location. The folder structure will look like this:

```
run_classifier/
├── _internal/          # internal dependencies (do not modify)
└── run_classifier.exe  # executable file
```

### 3. Run the `.exe`

Simply double-click the `.exe` file or launch it from the terminal:

```bash
.\run_classifier.exe
.\run_prediction.exe
.\run_trainer.exe
```

---

## Application Descriptions

**`run_classifier`** — a tool for manual or batch comment classification. Useful for quick review and labeling.

**`run_prediction`** — the main inference application. Takes comments as input and returns predicted classes.

**`run_trainer`** — fine-tunes the model on new data. Allows you to retrain the classifier on your own dataset.

---

## Environment Configuration

The repository includes a `.env` file with environment variables (e.g., file paths, parameters). Edit it as needed before running the applications.

---

## Source Code

The full source code (training, data labeling, scripts) is available on GitHub:
👉 [gerageragera39/Lubarsky_Comments_Model](https://github.com/gerageragera39/Lubarsky_Comments_Model)

Source repository structure:
- `data_hand_classifier/` — tools for manual data labeling
- `rubert_trainer/` — RuBERT fine-tuning scripts
- `dataset.csv` — main training dataset
- `test_comments.csv` — test set
- `result.png` — training results visualization

---

## Technical Details

- **Base model:** RuBERT (DeepPavlov)
- **Framework:** PyTorch + HuggingFace Transformers
- **Build:** PyInstaller (standalone Windows executables)
- **Data language:** Russian
- **Task:** Text Classification

---

## License

This project is released under the [MIT License](https://opensource.org/licenses/MIT).  
You are free to use, modify, and distribute this software for both personal and commercial purposes, provided that the original copyright notice is retained.

> **Note:** Since the applications are packaged with PyInstaller, they may be flagged by antivirus software as suspicious. This is a known false positive common to PyInstaller-built executables. You may need to add an exception in your antivirus or temporarily disable it to run the applications.

The software is provided **as is**, without warranty of any kind. The author takes no responsibility for any issues, damages, or data loss that may arise from its use.