snwfdhmp / llm

Use any LLM from the command line.

llm: language models usage made easy

Manipulate any language model from the command line.

From simple to advanced usage.

$ llm "Hello world."
Hello there! How can I assist you today?

Leave a ⭐ star to support the project.

Models

Some models are still being added. This is a work in progress.

Model Name	Status	Description
EVERY OpenAI model	✅
gpt-3.5-turbo	✅	ChatGPT
gpt-4	✅	GPT-4 via API (waitlist)
text-davinci-003	✅	InstructGPT (GPT-3)
llama2	✅	Meta's Llama 2
bing-chat	✅	Bing Chat: creative, balanced, precise
bert	✅	BERT by Google
llama-7b-hf	✅	Famous llama model
wizardlm-13b-uncensored	✅	WizardLM 30B
guanaco-65b-gptq	✅	Guanaco 65B
bard	🔄	Google Bard
... HuggingFace 🤗 models	✅	every text-generation model

Other models can be installed using the --install command.

Features

Feature	Status	Comment
Prompt	✅	Prompt model with default parameters
Parameterization	✅	temperature, max-length, top-p, top-k, ...
ChatGPT Plugins	🔄	Use chatGPT plugins. web-pilot working, global plugin system in development
Use files	✅	Query models using prompt files
Prompt chaining	✅	Call prompts like functions
Prompt templating	🔄	Use variables in prompt files

Getting started

git clone https://github.com/snwfdhmp/llm && cd llm
yarn install

make an alias llm

alias llm="node $(pwd)/main.js"

add it to your .bashrc or .zshrc to make it permanent.

You're ready to go ! Try:

$ llm "Hello world"
$ llm -m bing-creative "Tell me a joke"
$ llm -m gpt-3.5-turbo "Tell me a joke"

Usage

Simple prompting with defaults parameters

$ llm "what is the meaning of life?"

Use a specific model

$ llm -m bing-creative "find project ideas to learn react"

Use custom parameters

$ llm --max-length 512 --temperature 1 --top-p 0.9 --top-k 60 "follow the instructions."

List available models

$ llm ls
Name                LastUsedAt     Author      Description
text-davinci-003    2021-10-10     OpenAI      InstructGPT by OpenAI
gpt-3.5-turbo       2021-10-10     OpenAI      ChatGPT by OpenAI
gpt-4-web           2021-10-10     OpenAI      GPT-4 by OpenAI via chatGPT
llama               2021-10-10     Meta        Meta's Llama
bard                2021-10-10     Google      Google Bard
...

Use files as prompts

$ llm -f ./prompt.txt

Incoming:

• Conversation system (remember past messages)
• Install 3rd party models
• Chaining

$ llm -s session_name "what is the meaning of life?"
remembers past messages
$ llm --install github.com/snwfhdmp/llm-descriptor-llama
downloads model from github

Add any model

Any model can be plugged into llm using a model descriptor.

Example of a model descriptor which requires installation

kind: llm/descriptor/v1
metadata:
    name: llama
model:
    install: |
        git clone ...
        cd ...
        ./install.sh
        # or
        docker pull ...
        # or
        none
    usage:
        ./model-executor -f model.bin $LLM_PARAM_PROMPT
    parameters:
        LLM_PARAM_PROMPT:
            type: string
            description: The prompt to use
            default: "Hello world"
        LLM_PARAM_MAX_TOKENS:
            type: int
            description: The maximum length of context
            default: 100
        LLM_PARAM_TEMPERATURE:
            type: float
            description: The temperature of the model
            default: 0.7

Example of a model descriptor which uses an API

kind: llm/descriptor/v1
metadata:
    name: llama
model:
    install: |
        read -p "Enter your API key:" LLM_API_KEY
        echo "LLM_API_KEY=$LLM_API_KEY" >> ~/.bashrc
    usage: curl -s $LLM_PARAM_API_TARGET_URL -d "prompt=$LLM_PARAM_PROMPT&api_key=$LLM_API_KEY"
    parameters:
        LLM_PARAM_API_TARGET_URL:
            type: string
            description: The URL of the API
            default: "https://api.llm.com"
        LLM_PARAM_PROMPT:
            type: string
            description: The prompt to use
            default: "Hello world"
        LLM_PARAM_MAX_TOKENS:
            type: int
            description: The maximum length of context
            default: 100
        LLM_PARAM_TEMPERATURE:
            type: float
            description: The temperature of the model
            default: 0.7

Env variables

These variables can be used to tweak llm behavior.

• LLM_DEFAULT_MODEL - The default model to use when no model is specified
• LLM_ENABLED_PLUGINS - A comma-separated list of plugins to enable
• OPENAI_ORGANIZATION_ID - The organization ID to use for OpenAI models

Roadmap

Project vision and information can be found in docs.

Contributing

Contribute easily by leaving a ⭐ star to the project.

Code contributions are welcome. Please open an issue or a pull request.

Join the team at discord.gg/ccDghPeAT9.

llm's Issues

[DEP0040] DeprecationWarning: The `punycode` module is deprecated. Please use a userland alternative instead.

llm -m bing-creative "find project ideas to learn react":

(node:19496) [DEP0040] DeprecationWarning: The `punycode` module is deprecated. Please use a userland alternative instead.
(Use `node --trace-deprecation ...` to show where the warning was created)

And no other output

TypeError: openai.listModels is not a function

llm ls

List all models

Options:
  --help     Show help                                                 [boolean]
  --version  Show version number                                       [boolean]

TypeError: openai.listModels is not a function
    at Object.handler (file:///D:/AI/llm/main.js:100:36)
    at file:///D:/AI/llm/node_modules/yargs/build/lib/command.js:206:54
    at maybeAsyncResult (file:///D:/AI/llm/node_modules/yargs/build/lib/utils/maybe-async-result.js:9:15)
    at CommandInstance.handleValidationAndGetResult (file:///D:/AI/llm/node_modules/yargs/build/lib/command.js:205:25)
    at CommandInstance.applyMiddlewareAndGetResult (file:///D:/AI/llm/node_modules/yargs/build/lib/command.js:245:20)
    at CommandInstance.runCommand (file:///D:/AI/llm/node_modules/yargs/build/lib/command.js:128:20)
    at [runYargsParserAndExecuteCommands] (file:///D:/AI/llm/node_modules/yargs/build/lib/yargs-factory.js:1386:105)
    at YargsInstance.parse (file:///D:/AI/llm/node_modules/yargs/build/lib/yargs-factory.js:707:63)
    at file:///D:/AI/llm/main.js:123:4
    at ModuleJob.run (node:internal/modules/esm/module_job:218:25)