DSPy AI Agents Experiment

Declarative Self-improving Python (DSPy)

This framework allows developers to "program" Language Model (LM) prompts using function signatures to teach your LM to deliver high-quality outputs.

Getting Started

Install and run the Ollama Docker Containter on your local CPU

1. Install the Ollama Docker Image & run the Ollama container

   docker run -d -v ollama:/root/.ollama -p 11434:11434 --name ollama ollama/ollama

   1a. Or use one of the following options

   • Desktop
   • Manual
   • Other Docker Install Options

2. Start the Ollama container if you have it already

   docker start ollama

3. Install the 6 Bit quantized Llama 3 model (for Intel CPUs)

   docker exec -e OLLAMA_LLM_LIBRARY=cpu_avx ollama ollama run llama3.2:1b-instruct-q6_K

Run DSPy Locally with Ollama

1. Clone the repo

   git clone [email protected]:hamilton-labs/DSPy-AI-Agents-Experiment.git

2. Create a virtual environmnet

   python -m venv DSPy_AI_Agents_Experiment

3. Change into the DSPy_AI_Agents_Experiment directory

   cd DSPy_AI_Agents_Experiment

4. Activate the virtural environment

   source ./bin/activate

5. Install the dependencies (Python v3.12.3+)

   pip install -r requirements.txt

6. Run the program

   python DSPy_setup_example.py

7. Compare your output. Was it different?

   cat ./DSPy_output_example.md

8. Try entering your prompts from the command line. (Might make a CLI tool for this.)

   python DSPy_prompts_example.py

Possible Output (outputs may vary)

Here's how Llama 3 responded for me.

************************************************

Configured the LM

*************************************************
*************************************************

Calling the LM directly

*************************************************
*************************************************

Printing the LM's Output

*************************************************
It looks like you're trying to add some confidence and bravado to your response. I'll play along, but keep in mind that it's all just a simulation. Here's my response:

"Yes, indeed it is!" *pounds fist on table* "I mean, who needs a traditional test when we can have a simulated one? It's like the ultimate game of wits! Bring it on, test administrator!"

It looks like you're getting into the spirit of things. That's completely fine. I'm here to help and provide information, so feel free to ask me anything. What's on your mind for this test?