yamllm

A YAMLScript tool to query LLMs

Synopsis

$ yamllm 'Why pink?'
Q: Why pink?
llama3-8b-8192:
The question of pink!

The color pink has a long and complex history, and its associations and
meanings have evolved over time. Here are some possible reasons why pink might
be associated with certain qualities or themes:
...
$

Overview

This is a project to play with various LLMs from the command line.

At the moment it works with Groq which you can use for free with generous limits. It's also pretty fast.

It also works with OpenAI's API, which you have to pay for but costs pennies for single user interactive usage.

The tool is written in YAMLScript and requires a recent ys.

You can install it here (in $PWD/bin/) with make install-ys.

Installation

• Get a Groq API key here.
• Or get an OpenAI API key here.

Run these commands:

$ make install-ys     # Install YAMLScript `bin/ys`
$ export YAMLLM_GROQ_API_KEY=<your-groq-api-key>
$ # or: export YAMLLM_OPENAI_API_KEY=<your-openai-api-key>
$ source .rc          # Adds `$PWD/bin` to your PATH (for bash or zsh)
$ yamllm --help       # Ask it for some help
$ yamllm 'Why pink?'  # Get to the important stuff

These programs make it nicer to use but are not required:

• Install rlwrap for readline support and query history.
• Install prettier for 80 column output reformatting.

  npm install -g prettier
  

Try in a Browser with Github Codespaces

Use the Codespace Quickstart Link to try out yamllm in a browser environment.

CLI Usage

To start a query loop:

$ yamllm

Use ctl-d, exit or q to exit.

To run a single query:

$ yamllm 'Tell me a story'

Environment Variables

To use these variables, be sure to export them after setting them. One of the first 2 here is required.

• YAMLLM_GROQ_API_KEY=<groq-api-key>

  Required to access the Groq API (free). Get one here.

• YAMLLM_OPENAI_API_KEY=<open-api-key>

  Required to access the OpenAI API (not free, but not expensive). Get one here.

• YAMLLM_MODEL=<model-id>

  Choose one of the Groq or OpenAI models:

  * llama3    (llama3-70b-8192      Meta)
  * llama38b  (llama3-8b-8192       Meta)
  * gemma     (gemma-7b-it          Google)
  * gemma2    (gemma2-9b-it         Google)
  * mixtral   (mixtral-8x7b-32768   Mistral)
  * whisper   (whisper-large-v3     OpenAI)
  * gpt4o-    (gpt-4o-mini          OpenAI)
  * gpt4o     (gpt-4o               OpenAI)
  * gpt4t     (gpt-4-turbo          OpenAI)
  * gpt4      (gpt-4                OpenAI)
  * gpt3      (gpt-3.5-turbo        OpenAI)
  * dalle2    (dall-e-2             OpenAI)
  * dalle3    (dall-e-3             OpenAI)
  

  Default is llama3 if YAMLLM_GROQ_API_KEY is set, gpt4o- if YAMLLM_OPENAI_API_KEY is set else error.

• YAMLLM_LOG_DIR=<directory-path>

  A directory to store query results. Directory can be relative or absolute but must exist.

• YAMLLM_PRE=<file>

  File containing text to prepend to every prompt sent to API.

• YAMLLM_POST=<file>

  File containing text to append to every prompt sent to API.

• YAMLLM_PROMPT=<file>

  File containing the prompt text to send to API. YAMLLM_PRE and YAMLLM_POST are added if specified.

• YAMLLM_IMAGE_SIZE=<size-indicator>

  One of 1024x1024, 1024x1792 or 1792x1024.

• YAMLLM_TEMP=<number 0.0 - 2.0>

  Higher is more random. Lower is more deterministic. Default is 0.8.

• YAMLLM_TOPP=<number 0.0 - 1.0>

  Alternative way to do something like YAMLLM_TEMP. Default is 1.0.

• YAMLLM_TEST_FILE=$PWD/test/test1/yaml

  Set this to run the YAML interaction test suite.

• YAMLLM_DEBUG=1

  Dump API payloads.