applemcg / salamander

Salamander - a simple LLM front end

Salamander - a simple LLM front end

2024-03-03:

• forked this repo from Peter Kaminski
• expanded to include different LLM models

Python programs:

- these Python programs use the direct REST interface provided by
  the PyPI `requests` module and do not require the `pip`
  installation of LLM specific modules.  
  
- each LLM does require an API key  

- `-h` or `--help` provides argument descriptions

• salamander.py is the original program written to provide OpenAI text output in response to an input file that contains some prompt text.

• gecko.py is a similar program that uses Anthropic's Claude-3-sonnet model. Specify a model name ("opus", "sonnet", "haiku") with the -m argument.

• iguana.py uses Google's Gemini-Pro model.

How to use these programs:

• One way to use these programs is to set up a local Python virtual environment and run them in the terminal:
  • initial setup:

python3 -m venv venv
source venv/bin/activate
pip install -U pip
pip install -r requirements.txt

• My working practice is to activate the virtual environment
  ($ source venv/bin/activate)
  before running any code and deactivating ($ deactivate) when done.

• run these programs (using a simple "why is the sky blue?" prompt):

  • first set up API keys: see env.sh-template for an example of how to do this.

./salamander.py -i prompts/skybluePrompt.md
./gecko.py -i prompts/skybluePrompt.md
./iguana.py -i prompts/skybluePrompt.md

• The code generation prompt that was fed to "ChatGPT Plus" to create the first sample of salamander.py is found in prompts/salamanderPrompt.md. It can be used to compare how OpenAI, Claude, and Gemini respond to a prompt like this.

Plans and TODOs:

• experiment with some summarization and topic extraction prompts

  • two examples in the prompts/ directory

• experiment with using the previous response as part of a topic and summarization conversation

Please post issues.