Tunaas-r1:32b

A customized version of DeepSeek-R1 32B model with enhanced conversational abilities and optimized parameters.

Model Information

Base Model: DeepSeek-R1 32B
Size: 19 GB
Quantization: Q4_K_M
Context Length: 128,000 tokens
License: Same as DeepSeek-R1

Installation

To use this model, you need to have Ollama installed on your system.

ollama run neuroequality/tunaas-r1:32b

Features

Custom System Prompt

Tunaas-r1 has been fine-tuned with a custom system prompt that emphasizes:

Helpful and thoughtful responses
Detailed explanations when needed
Concise answers when appropriate
Honest and harmless interactions

Optimized Parameters

Temperature: 0.7 (balanced creativity and coherence)
Top-p: 0.9 (nucleus sampling for quality)
Top-k: 40 (token selection diversity)
Max Tokens: 128,000

Chat Template

Uses DeepSeek’s optimized chat template format for better conversation flow:

<|im_start|>system
{system_message}
<|im_end|>
<|im_start|>user
{user_message}
<|im_end|>
<|im_start|>assistant
{assistant_response}

Usage Examples

Basic Usage

ollama run neuroequality/tunaas-r1:32b

API Usage

curl http://localhost:11434/api/generate -d '{
  "model": "neuroequality/tunaas-r1:32b",
  "prompt": "Explain quantum computing in simple terms"
}'

Python Example

import requests
import json

response = requests.post('http://localhost:11434/api/generate', 
    json={
        "model": "neuroequality/tunaas-r1:32b",
        "prompt": "Write a Python function to calculate fibonacci numbers",
        "stream": False
    })

print(json.loads(response.text)['response'])

Use Cases

Tunaas-r1:32b excels at:

Code Generation: Writing and explaining code in multiple languages
Creative Writing: Stories, articles, and content creation
Technical Explanations: Breaking down complex topics
Problem Solving: Step-by-step reasoning and analysis
General Conversation: Engaging and helpful dialogue

Performance

Hardware Requirements

Minimum RAM: 20 GB
Recommended RAM: 24 GB or more
GPU: Optional but recommended for faster inference

Benchmarks

The model maintains the strong performance characteristics of DeepSeek-R1:

Excellent reasoning capabilities
Strong code generation
Nuanced language understanding
Multilingual support

Customization

You can further customize this model by creating your own Modelfile:

FROM neuroequality/tunaas-r1:32b

# Add your custom parameters
PARAMETER temperature 0.8
PARAMETER top_p 0.95

# Add your custom system message
SYSTEM "Your custom system prompt here"

Then create your variant:

ollama create my-custom-model -f ./Modelfile

Model Card

Intended Use

General-purpose AI assistant
Educational and research purposes
Creative and technical writing
Code development assistance

Limitations

May produce incorrect or biased information
Should not be used for critical decision-making without verification
Responses should be reviewed for accuracy
Not suitable for generating harmful content

Ethical Considerations

This model is designed to be helpful, harmless, and honest. It includes safety measures from the base DeepSeek-R1 model and additional customizations to promote positive interactions.

Credits

Base Model: DeepSeek-R1 by DeepSeek AI
Customization: NeuroEquality
Platform: Ollama

License

This model inherits the license from DeepSeek-R1. Please refer to the original model’s license for usage terms.

Support

For issues or questions:

Open an issue on the Ollama model page
Visit the Ollama Discord
Check the Ollama documentation

Changelog

Version 1.0 (Initial Release)

Based on DeepSeek-R1 32B
Custom system prompt for enhanced helpfulness
Optimized generation parameters
Custom chat template implementation

Made with ❤️ by NeuroEquality

Enhanced Reasoning Assistant optimized for helpful, detailed, and honest conversations.

Readme