README.md

---
language:
- en
license: apache-2.0
library_name: transformers
tags:
- tokenizer
- rx-codex
- medical-ai
- code-tokenizer
- chat-ai
pipeline_tag: text-generation
---
<div align="center">
  <img src="./rx_codex_logo.png" width="50%" alt="Rx-Codex-AI" />
</div>
<h3 align="center">
  <b>
    <span>━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━</span>
    <br/>
    Rx Codex Tokenizer: Professional Tokenizer for Modern AI
    <br/>
    <span>━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━</span>
    <br/>
  </b>
</h3>
<br/>

<div align="center" style="line-height: 1;">
  |
  <a href="https://huggingface.co/rxmha125" target="_blank">🤗 HuggingFace</a>
  &nbsp;|
  <a href="https://rxcodexai.com" target="_blank">🌐 Website</a>
  &nbsp;|
  <a href="mailto:contact@rxcodexai.com" target="_blank">📧 Contact</a>
  &nbsp;|

  <br/>
</div>

<br/>

## Overview

Rx Codex Tokenizer is a state-of-the-art BPE tokenizer designed for modern AI applications. With 128K vocabulary optimized for English, code, and medical text, it outperforms established tokenizers in comprehensive benchmarks.

**Developed by Rx Founder & CEO of Rx Codex AI**

## Benchmark Results

### Tokenizer Battle Royale - Final Scores

| Tokenizer | Final Score | Speed | Compression | Special Tokens | Chat Support |
|-----------|-------------|-------|-------------|----------------|--------------|
| 🥇 **Rx Codex** | **84.51/100** | 24.84/25 | 35.0/35 | 16.67/20 | 15/15 |
| 🥈 GPT-2 | 67.89/100 | 24.89/25 | 35.0/35 | 0.0/20 | 15/15 |
| 🥉 DeepSeek | 67.77/100 | 24.77/25 | 35.0/35 | 0.0/20 | 15/15 |

### Final Scores Comparison
![Final Scores](./final_scores_comparison.png)

### Speed Analysis
![Speed Comparison](./speed_comparison.png)

### Compression Efficiency
![Compression Comparison](./compression_comparison.png)

### Multi-dimensional Analysis
![Capabilities Radar](./capabilities_radar.png)

### Token Count Efficiency
![Token Count Comparison](./token_count_comparison.png)

## Key Features

- **128K Vocabulary** - Optimal balance of coverage and efficiency
- **Byte-Level BPE** - No UNK tokens, handles any text
- **Medical Text Optimized** - Perfect for healthcare AI applications
- **Code-Aware** - Excellent programming language support
- **Chat-Ready Tokens** - Built-in support for conversation formats

## Technical Specifications

- **Vocabulary Size**: 128,256 tokens
- **Special Tokens**: 9 custom tokens
- **Model Type**: BPE with byte fallback
- **Training Data**: OpenOrca 5GB English dataset
- **Average Speed**: 0.63ms per tokenization
- **Compression Ratio**: 4.18 characters per token

## Use Cases

- **Chat AI Systems** - Built-in chat token support
- **Medical AI** - Optimized for healthcare terminology
- **Code Generation** - Excellent programming language handling
- **Academic Research** - Efficient with complex text

## License

Apache 2.0

## Author

**Rx Founder & CEO**  
Rx Codex AI