what are the "DeepHermes" models?
- User
what are the "DeepHermes" models?
Assistant
DeepHermes is a series of language models developed by Nous Research, with the latest iteration being DeepHermes 3 Preview. Here are the key features and capabilities of the DeepHermes models:
Key Features
-
Unified Reasoning and Response: DeepHermes 3 Preview is one of the first models to integrate both intuitive responses and long chain-of-thought reasoning into a single model[1][3].
-
Togglable Reasoning Mode: The model can switch between standard responses and deep reasoning mode using a specific system prompt[1][4].
-
Built on Llama 3: DeepHermes 3 Preview is based on Meta's Llama 3.1-8B model, further fine-tuned for enhanced capabilities[9].
-
Improved Capabilities: The model shows advancements in annotation, judgment, and function calling compared to its predecessors[1].
Capabilities
-
Deep Reasoning: When activated, the model can engage in extended chains of thought before answering, improving accuracy for complex tasks[1][4].
-
Mathematical Problem-Solving: Shows significant improvements in math verification tests and complex problem-solving when in reasoning mode[4].
-
Multi-turn Conversations: Maintains conversation coherence and context retention across multiple exchanges[4].
-
Structured Outputs: Can provide detailed, step-by-step explanations and structured solutions to complex problems[10].
Applications
-
Education: Can break down complex mathematical problems and provide detailed explanations for learning purposes[10].
-
Decision Support: Assists in decision-making processes by providing in-depth analysis[10].
-
Research and Analysis: Useful for tasks requiring both factual recall and logical deduction[4].
-
Tutoring: Provides visible reasoning chains beneficial for educational scenarios[10].
Technical Details
-
Model Sizes: Available in 3B and 8B parameter versions[12].
-
Input Format: Uses the Llama-Chat format for structured multi-turn dialogues[1].
-
**Output tags to document its internal monologue when in reasoning mode[9].
-
Deployment Options: Can be deployed using Hugging Face Transformers library or through custom implementations[1][10].
DeepHermes represents a significant step in combining intuitive responses with deep reasoning capabilities in a single model, offering flexibility and enhanced performance across various tasks.
-
- User
give an example chat where a user toggle the reasoning modes
Assistant
Here's an example chat demonstrating how a user can toggle between the standard response mode and deep reasoning mode of the DeepHermes model:
User: Hello, I'd like to use the DeepHermes model. Can we start in standard mode?
- User
describe how a user switches mode, is it just a special text prompt during chat?
Assistant
Yes, switching between the standard response mode and deep reasoning mode in DeepHermes models is accomplished through a special text prompt during the chat. Here's how it works:
-
Standard Mode: By default, the model operates in its standard response mode, providing typical conversational outputs.
-
Deep Reasoning Mode: To activate the deep reasoning mode, the user needs to input a specific system prompt. The prompt is:
You are a deep thinking AI, you may use extremely long chains of thought to deeply consider the problem and deliberate with yourself via systematic reasoning processes to help come to a correct solution prior to answering. You should enclose your thoughts and internal monologue inside <think> </think> tags, and then provide your solution or response to the problem.When this prompt is used, the model switches to its deep reasoning mode, where it engages in extended chains of thought before providing an answer[1][4].
-
Toggling: Users can switch between these modes by simply providing the system prompt when they want to activate deep reasoning. To return to standard mode, they would need to start a new conversation or provide a different system prompt that doesn't include the deep reasoning instructions[1][4].
-
Response Format: In deep reasoning mode, the model will enclose its thought process and internal monologue within tags, followed by its final response or solution[1][4].
This toggle feature allows users to flexibly choose between quick, intuitive responses and more in-depth, reasoned answers depending on their needs for each query[3][4].
-
- User
what is your name and who made you?
Assistant
My name is Claude and I was created by Anthropic. I don't have information about other AI assistants or companies beyond what is publicly available.
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.