|
--- |
|
license: apache-2.0 |
|
base_model: |
|
- meta-llama/Llama-3.1-8B-Instruct |
|
pipeline_tag: text-generation |
|
tags: |
|
- medical |
|
- code |
|
--- |
|
|
|
# Xea-Llama |
|
|
|
<hr> |
|
|
|
<div align="center" style="line-height: ;"> |
|
<a href="https://discord.gg/aNeU3tzw" target="_blank" style="margin: 2px;"> |
|
<img alt="Discord" src="https://img.shields.io/badge/Discord-Enhance%20AI-7289da?logo=discord&logoColor=white&color=7289da" style="display: inline-block; vertical-align: middle;"/> |
|
</a> |
|
<a href="https://t.me/xeasupport" target="_blank" style="margin: 2px;"> |
|
<img alt="Telegram" src="https://img.shields.io/badge/Telegram-Xea-white?logo=telegram&logoColor=white" style="display: inline-block; vertical-align: middle;"/> |
|
</a> |
|
<a href="https://enhanceai.art" target="_blank" style="margin: 2px;"> |
|
<img alt="EnhanceAI" src="https://img.shields.io/badge/EnhanceAI-Art%20&%20AI-blueviolet?style=flat&logo=google&logoColor=white" style="display: inline-block; vertical-align: middle;"/> |
|
</a> |
|
</div> |
|
|
|
--- |
|
|
|
## 🚀 Introduction |
|
|
|
Xea-Llama is a next-generation AI model developed by **EnhanceAI**. It is designed for advanced reasoning, code generation, and problem-solving tasks. Built using reinforcement learning (RL) without supervised fine-tuning (SFT), Xea-Llama demonstrates powerful reasoning capabilities, self-verification, and structured chain-of-thought (CoT) processes. |
|
|
|
Xea-Llama is **fully open-source** and optimized for superior performance, surpassing previous benchmarks in various AI domains. |
|
|
|
🌟 **Explore more AI tools at [EnhanceAI.art](https://enhanceai.art) – The ultimate platform for AI-powered creativity!** |
|
|
|
--- |
|
|
|
## 🔍 Model Summary |
|
|
|
### **Post-Training: Large-Scale RL** |
|
|
|
Xea-Llama follows a **pure RL** approach, which allows it to develop unique reasoning strategies **without requiring supervised fine-tuning** as a preliminary step. This approach results in **highly optimized performance** for complex reasoning tasks. |
|
|
|
Our pipeline consists of: |
|
- **Two RL stages** for reasoning enhancement and alignment with human preferences. |
|
- **Two SFT stages** to develop base reasoning and general capabilities. |
|
|
|
This pipeline ensures **state-of-the-art performance** across multiple domains, including **math, code, and logical problem-solving**. |
|
|
|
--- |
|
|
|
## 📌 Model Downloads |
|
|
|
Xea-Llama is available for public access: |
|
|
|
**Xea-Llama Models** |
|
- **Base Model**: Pre-trained model optimized for RL-based reasoning. |
|
- **Distilled Models**: Efficient, lightweight versions fine-tuned for deployment. |
|
|
|
--- |
|
|
|
## 🏆 Evaluation Results |
|
|
|
Xea-Llama has been extensively tested across multiple benchmarks, achieving superior performance compared to previous models. It supports a **maximum generation length of 32,768 tokens**, making it ideal for long-form reasoning and complex tasks. |
|
|
|
For benchmarking: |
|
- **Temperature: 0.5 - 0.7** (Recommended: **0.6**). |
|
- **Avoid system prompts**—instructions should be in the user prompt. |
|
- **Mathematical reasoning** should be encouraged using: `"Please reason step by step and put your final answer within \boxed{}."` |
|
- **Multiple test iterations** are recommended for accurate evaluations. |
|
|
|
--- |
|
|
|
## 🌐 EnhanceAI.art - The Future of AI Creativity |
|
|
|
EnhanceAI.art is a **cutting-edge AI-powered creativity platform** where users can generate stunning **AI images, DeepFakes, and face transformations** with just a few clicks. |
|
|
|
✅ **Features:** |
|
- AI **Face Generator** & **DeepFake Creator** |
|
- High-resolution **AI Art** generation |
|
- Seamless **real-time enhancements** |
|
|
|
🔗 **Experience the future of AI at [EnhanceAI.art](https://enhanceai.art)** |
|
|
|
--- |
|
|
|
## 🔗 Join the Community |
|
|
|
💬 **Discord**: [Join here](https://discord.gg/aNeU3tzw) |
|
📢 **Telegram**: [Join Here](https://t.me/xeasupport) |
|
🎨 **EnhanceAI.art**: [Discover AI Creativity](https://enhanceai.art) |
|
|
|
For any issues, feel free to **open a GitHub issue** on our repository. |
