DeepHermes-ToolCalling-Specialist-Atropos Open-Source Model - Enhance Inference Tool Calling Performance, Free to Use!

Deephermes ToolCalling Specialist Atropos

Developed by NousResearch

An experimental model fine-tuned by Nous Research using the Atropos reinforcement learning framework, focused on improving the tool calling performance of the Llama-3.1 8B model in inference mode

Large Language Model

Transformers

English#Reinforcement Learning Fine-tuning #Inference Mode Tool Calling #Parallel Function Execution

Downloads 64

Release Time : 4/11/2025

Model Overview

Based on the DeepHermes 3 Llama-3.1 8B model, it specifically optimizes tool calling capabilities through the Atropos reinforcement learning framework, making it particularly suitable for tool usage scenarios in complex reasoning tasks

Model Features

Reinforcement Learning Optimized Tool Calling

Significantly improves the accuracy of tool calling in inference mode through the Atropos reinforcement learning framework

Open Source Reinforcement Learning Framework

Developed based on the fully open-source Atropos reinforcement learning environment stack

Multi-mode Reasoning Support

Supports deep reasoning mode, standard dialogue/instruction mode, structured JSON output, and function calling

Model Capabilities

Complex reasoning task processing

Parallel tool calling

Structured JSON output generation

Function calling execution

Instruction following

Use Cases

Tool Calling Optimization

Parallel Function Calling

Handling multiple function call requests simultaneously

Accuracy improved from 0.10 to 0.46

Simple Tool Calling

Basic tool calling tasks

Accuracy improved from 0.21 to 0.5175

🚀 DeepHermes Tool Calling Specialist - Atropos RL

The DeepHermes Tool Calling Specialist - Atropos RL model is an experimental fine - tuned model that significantly enhances the tool - calling performance of the base model during reasoning mode.

🚀 Quick Start

The DeepHermes Tool Calling Specialist - Atropos RL model is an experimental artifact fine - tuned by Nous Research using our innovative open - source reinforcement learning framework—Atropos. This variant specifically improves the tool calling performance of the DeepHermes 3 Llama - 3.1 8B model during its reasoning mode.

⚠️ Important Note

This model is intended as an experimental artifact and is not designed for broad, general - purpose use.

✨ Features

Improved Tool Calling in Reasoning Mode: Reinforcement learning significantly boosts tool usage during complex reasoning tasks.
Open - Source RL Framework: Utilizes the fully open - source Atropos RL Environments.
Active Open Source Community: Contributions welcomed on the Atropos GitHub.
Upcoming SOTA RL Trainer: A state - of - the - art open - source reinforcement learning trainer by Nous Research is coming soon.

📦 Installation

No installation steps are provided in the original document, so this section is skipped.

💻 Usage Examples

This model supports multiple inference modes including:

Reasoning (Deep Thinking Mode)
Standard Chat/Instruction Mode
Structured JSON Outputs
Function Calling

Note: You must first place DeepHermes' reasoning system prompt, and then append your function calling system prompt after for it to do reasoning and tool calling simultaneously.

🌐 Hermes Function Calling GitHub

📚 Documentation

Atropos Open Source Framework

Atropos is Nous Research’s open - source Reinforcement Learning environment stack, designed to enhance various aspects of LLM functionalities through structured RL methodologies. We encourage contributions and exploration:

🌐 Atropos GitHub Repository

Benchmark Results

Evaluations on the Berkeley Function Calling benchmark demonstrate significant improvements in tool calling accuracy during reasoning mode, compared to its base model:

Benchmark	Base Accuracy	Atropos RL Accuracy	Improvement
Parallel	0.10	0.46	4.6x
Simple	0.21	0.5175	2.5x

These enhancements are due to RL fine - tuning specifically targeted at improving reasoning - based tool calling capabilities.

Eval set accuracy results:

![image/png](https://cdn - uploads.huggingface.co/production/uploads/6317aade83d8d2fd903192d9/BWIFtF7oBWO2b_Q1oh4YM.png)

🔧 Technical Details

No specific technical details beyond what's already covered are provided in the original document, so this section is skipped.

📄 License

This model is under the llama3 license.

📋 Information Table

Property	Details
Model Type	DeepHermes Tool Calling Specialist - Atropos RL
Base Model	meta - llama/Meta - Llama - 3.1 - 8B
Library Name	transformers
Tags	Llama - 3, RL, Atropos, Tool Calling, Nous Research, instruct, finetune, reasoning, function calling, transformers, reinforcement - learning, json mode, chatml

📚 Citation

@misc{
      title={DeepHermes Tool Calling Specialist - Atropos RL},
      author={Teknium and Dakota Mahan and Roger Jin and Chen Guang and Jai Suphavadeeprasit and Jeffrey Quesnelle},
      year={2025},
      url={https://huggingface.co/NousResearch/DeepHermes-Tool-Calling-Specialist-Atropos-RL}
}

👥 Community and Support

For questions, issues, or findings, please open issues or discussions in the respective GitHub repositories:

Nous Research encourages active community engagement and open - source contributions to continuously improve model performance and capabilities.

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご