#language-model #nlp #command-line-interface #llm #text #generate #bedrock

bin+lib hiramu-cli

A command-line interface tool for interacting with large language models (LLMs) on AWS Bedrock and generating text based on prompts

4 releases

0.1.23 Apr 23, 2024
0.1.16 Apr 21, 2024

#315 in Machine learning

Download history 5/week @ 2024-09-23

287 downloads per month

Apache-2.0

46KB
489 lines

Hiramu CLI

Hiramu CLI is a powerful command-line interface for interacting with language models. It provides a seamless way to ask questions and generate text using various models from different providers, including Anthropic's Claude (Haiku, Sonnet, Opus), Mistral (7B, 8x7B, Large), and Ollama.

Table of Contents

Features

  • Ask questions to language models using a simple command-line interface
  • Support for multiple models from Anthropic, Mistral, and Ollama
  • Customizable options for region, profile, max tokens, temperature, and model alias
  • Interactive input using the {input} placeholder in prompts
  • Real-time streaming of generated text
  • Integration with other command-line tools and databases (e.g., git diff, DuckDB, PostgreSQL)

Installation

To install Hiramu CLI, ensure you have Rust installed on your system. If you don't have Rust installed, you can follow the official installation guide: https://www.rust-lang.org/tools/install

Once Rust is installed, run the following command to install Hiramu CLI:

cargo install hiramu-cli

Usage

To ask a question to a language model, use the generate command followed by the question. You can specify additional options to customize the behavior of the CLI.

hiramu-cli generate "What is the capital of France?" -r us-west-2 -p bedrock -m 100 -t 0.7 -M haiku -P bedrock

Options

  • -r, --region <REGION>: The region to use (default: "us-west-2").
  • -p, --profile <PROFILE>: The profile to use (default: "bedrock").
  • -m, --maxtoken <MAXTOKEN>: The maximum number of tokens to generate (default: 100).
  • -t, --temperature <TEMPERATURE>: The temperature for generation (default: 0.7).
  • -M, --model <MODEL>: The model alias to use (default: "haiku").
  • -P, --provider <PROVIDER>: The provider alias to use for generation (default: "bedrock").
  • -E, --endpoint <ENDPOINT>: The provider endpoint to use for generation (default: "http://localhost:11434").

Interactive Input

Hiramu CLI supports interactive input using the {input} placeholder in prompts. When the placeholder is present, the CLI will prompt you to enter the input, which will be inserted into the prompt before sending it to the language model.

hiramu-cli generate "Translate the following text from English to French: {input}" -M sonnet

This feature allows you to provide dynamic input to the language model during runtime.

Model Aliases

Hiramu CLI provides convenient aliases for different language models. The following model aliases are available:

  • haiku: Anthropic Claude 3, Haiku 1x
  • sonnet: Anthropic Claude 3, Sonnet 1x
  • opus: Anthropic Claude 3, Opus 1x
  • mistral7b: Mistral 7B Instruct 0x
  • mistral8x7b: Mistral 8x7B Instruct 0x
  • mistral-large: Mistral Large

You can use these aliases with the -M or --model option to specify the desired model for generation.

Provider Aliases

Hiramu CLI supports different providers for language model generation. The following provider aliases are available:

  • bedrock: Anthropic's Bedrock platform (default)
  • ollama: Ollama provider

You can use these aliases with the -P or --provider option to specify the desired provider for generation. When using the ollama provider, you also need to specify the endpoint using the -E or --endpoint option.

Examples

Here are a few examples demonstrating the usage of Hiramu CLI:

  1. Ask a question using the default options:

    hiramu-cli generate "What is the capital of France?"
    
  2. Specify model and temperature:

    hiramu-cli generate "What is the meaning of life?" -M sonnet -t 0.5
    
  3. Translate interactively:

    hiramu-cli generate "Translate from English to Spanish: {input}" -M mistral8x7b
    
  4. Generate release notes by combining with git diff:

    git diff HEAD~1..HEAD | hiramu-cli generate "Summarize the changes:" -M opus
    

    This pipes the output of git diff into Hiramu CLI to generate a summary of the code changes.

  5. Generate SQL queries from natural language using DuckDB:

    hiramu-cli generate "SQL query to find users who signed up in the last 30 days: {input}" -M mistral7b | duckdb -c -
    

    The generated SQL query is piped directly into DuckDB for execution.

  6. Optimize SQL queries using PostgreSQL:

    query="SELECT * FROM orders JOIN customers ON orders.customer_id = customers.id"
    optimized_query=$(echo "$query" | hiramu-cli generate "Optimize this SQL query:" -M mistral-large)
    psql -d mydb -c "$optimized_query"
    

    The existing SQL query is passed to Hiramu CLI to generate an optimized version, which is then executed using PostgreSQL.

Feel free to explore different prompts, models, providers, and options to generate various types of content using Hiramu CLI.

Contributing

Contributions to Hiramu CLI are welcome! If you encounter any issues or have suggestions for improvements, please open an issue on the GitHub repository.

Before submitting a pull request, ensure that the tests pass and the code is formatted with cargo fmt. You can run the tests using the following command:

hiramu-cli generate "Once upon a time, in a far-off land, there lived a brave knight named {input}. The knight embarked on a quest to..." -m 200 -M mistral-large

Feel free to explore different prompts, models, and options to generate various types of content using Hiramu CLI.

Contributing

Contributions to Hiramu CLI are welcome If you encounter any issues or have suggestions for improvements, please open an issue on the GitHub repository.

Version

0.1.21

  • Fix issue with {input} handling

0.1.16

  • Replace prompt command with generate
  • Support for temperature and maxtoken for Ollama

License

Hiramu CLI is open-source software licensed under the Apache 2 License. See the LICENSE file for more details.

Dependencies

~24–37MB
~536K SLoC