Interacting with Large Language Models via 'llm' Command (with Examples)

The llm command is a powerful tool that allows users to interact seamlessly with large language models (LLMs) through remote APIs and locally installed models. It serves as a versatile interface, enabling users to execute a wide range of tasks, from running basic prompts to engaging in interactive chats with models. With llm, users can efficiently navigate and utilize the expansive capabilities of language models, whether for educational, professional, or creative purposes.

Use Case 1: Set up an OpenAI API Key

Code:

llm keys set openai

Motivation:

Setting up an API key is a critical initial step for users who want to leverage the power of OpenAI’s models. This command is motivated by the need to authenticate and authorize access to OpenAI’s services. Without this setup, users cannot send requests or receive results from the large language models provided by OpenAI.

Explanation:

• llm: This is the base command used to interact with large language models.
• keys: This argument indicates that the command is dealing with API keys.
• set: This specifies the action of setting or configuring an API key.
• openai: This refers to the specific service provider, OpenAI, whose API is being accessed.

Example Output:

Upon running this command, a prompt will likely appear, requesting the user to input their OpenAI API key. After successful entry, a confirmation message will indicate that the API key has been set.

Use Case 2: Run a Prompt

Code:

llm "Ten fun names for a pet pelican"

Motivation:

Running prompts is fundamental for generating creative outputs or obtaining data-driven insights. In this example, the motivation is to receive amusing and inventive name suggestions for a pet pelican, showcasing how the model can assist with creative brainstorming.

Explanation:

• llm: The core command for engaging with language models.
• The text "Ten fun names for a pet pelican" is the actual prompt. It queries the language model to apply its creativity and linguistic capabilities to generate a list of playful names.

Example Output:

Output could be a list such as:

1. Percy the Pelican
2. Beaky McBeakface
3. Splashy Pete
4. Flapjack
5. Captain Pelly
6. Sir Squawksalot
7. Wingnut
8. Pelicawesome
9. Gulliver
10. Fishy Fred

Use Case 3: Run a System Prompt Against a File

Code:

cat path/to/file.py | llm --system "Explain this code"

Motivation:

This use case is motivated by the need to understand existing code. Whether examining legacy code or exploring unfamiliar scripts, leveraging a language model to explain code can significantly aid developers and students in grasping functionality, improving code comprehension and documentation processes.

Explanation:

• cat path/to/file.py: This command outputs the content of a file. Replace path/to/file.py with the actual path to the Python file to be analyzed.
• |: The pipe operator sends the output from the cat command to the llm command.
• llm: The base command for engaging with language models.
• --system: This flag indicates a system-level prompt, designed for tasks such as code explanation.
• "Explain this code": The system prompt that asks the language model to describe and elucidate the function and logic of the code in the given file.

Example Output:

The response could be a detailed explanation of each function and key statement within the file, helping the user to understand the code’s logic and purpose.

Use Case 4: Install Packages from PyPI into the Same Environment as LLM

Code:

llm install package1 package2 ...

Motivation:

Installing additional packages can extend the functionality of the llm environment. This is particularly crucial when a user requires specific libraries for custom data processing or model manipulation tasks. The motivation here is to ensure that all necessary tools are readily available within the llm environment, streamlining workflows.

Explanation:

• llm: The foundational command interface for interacting with LLMs.
• install: This command tells the system to fetch and install software packages.
• package1 package2 ...: Replace package1, package2 with the actual package names you wish to install. This allows for customization of the llm environment with additional Python packages available on PyPI.

Example Output:

The terminal will display a series of installation messages, confirming the successful installation of the specified packages.

Use Case 5: Download and Run a Prompt Against a Model

Code:

llm --model orca-mini-3b-gguf2-q4_0 "What is the capital of France?"

Motivation:

This use case highlights the ability to choose specific models for running prompts, providing flexibility based on performance and output needs. For instance, a user may opt for the orca-mini-3b model to leverage its unique characteristics for geographical questions or particular datasets.

Explanation:

• llm: The command used to interact with language models.
• --model: This option specifies that a particular model should be used.
• orca-mini-3b-gguf2-q4_0: Indicates the exact model to be used, which might be optimized for specific tasks or performance characteristics.
• "What is the capital of France?": The user-provided prompt requesting factual information from the language model.

Example Output:

The model would return the capital of France, which is “Paris”.

Use Case 6: Create a System Prompt and Save it with a Template Name

Code:

llm --system 'You are a sentient cheesecake' --save sentient_cheesecake

Motivation:

Saving system prompts as templates is motivated by the need for efficiency and reusability. Users who frequently interact with LLMs via specific system prompts benefit from creating templates, which streamline repeated tasks by saving time and reducing the possibility of error or omission.

Explanation:

• llm: The baseline command for interacting with LLMs.
• --system: This indicates that a system-level prompt is being used.
• 'You are a sentient cheesecake': The system prompt to be sent to the model, used here for creative exercises or entertainment.
• --save: Specifies that the following text should be the name of the saved template.
• sentient_cheesecake: The name given to this specific system prompt template for future reference.

Example Output:

A confirmation message that the template was saved successfully, allowing the user to easily reuse this system prompt in future interactions.

Use Case 7: Have an Interactive Chat with a Specific Model Using a Specific Template

Code:

llm chat --model chatgpt --template sentient_cheesecake

Motivation:

Interactive chat interfaces are increasingly important for dynamic and ongoing engagements, whether for support, learning, or companionship scenarios. Leveraging specific models with tailor-made templates enhances the quality, relevance, and user satisfaction of these interactions. This use case allows users to have a continuous conversation with a language model using a predefined character or situation.

Explanation:

• llm: Command that provides the interface for interacting with language models.
• chat: Initiates an interactive session with a language model.
• --model chatgpt: Specifies the model to be used, in this case, OpenAI’s ChatGPT, known for its conversational capabilities.
• --template sentient_cheesecake: Utilizes the previously saved template, which sets the context for the chat, adding a creative or thematic dimension to the interaction.

Example Output:

The session would begin with the model engaging as a “sentient cheesecake,” leading to a conversation that explores this whimsical scenario through model-generated responses.

Conclusion:

The llm command serves as a powerful and adaptable interface for interacting with large language models, supporting a wide range of tasks from basic queries to intricate system-level prompts and interactive sessions. It offers users the flexibility to exploit models both remotely and locally, enriching their engagement with advanced AI technologies through streamlined command-line interactions. These diverse use cases illustrate the broad utility of llm, accommodating creative, educational, technical, and exploratory roles.