Deeper Hermes

Deephermes is really cool because a single model can act either as a normal agent, or a deep thinking AI simply by using system prompts.

What if you don't even need a system prompt, and the model automatically acts based on its OWN judgement?

In above example:

1. The first question is a simple question, so the model acts as a basic agent
2. The second question is also simple, so it acts as a simple agent
3. The third question is a bit complex and requires resasoning, so it starts with [THINKING] and then ends with [ANSWER], confirming that it's using the deep thinking mode.

How it works

1. Deeper Hermes is made up of 2 separate queries.
2. The first one is to ask which mode to use (normal agent vs. deep thinking AI)
3. The answer from 2 decides the system prompt that gets fed into the actual question.

Get started

A. 1-click Launch

1. First install LM Studio
2. Install and Run Deeper Hermes on https://pinokio.computer with 1 click.

B. Manual Install

First install LM Studio

Then load nousresearch_deephermes-3-llama-3-8b-preview into LMStudio:

lms load nousresearch_deephermes-3-llama-3-8b-preview -y

Once the LMStudio server is running and has loaded the model, activate virtual environment

source env/bin/activate (mac/linux)
.\env\Scripts\activate (windows)

Next, install dependencies:

pip install -r requirements.txt

Then run with:

python app.py