Chatways

English | 中文

Chatways provides a simple way to build chat applications with LLMs, offering templates, interfaces, and support for multiple LLM backends.

Features

• Template-driven application creation for non-coders, enabling app launching with a single command.
• Simplified API interfaces for developers, eliminating complex development workflows.
• Integration with multiple LLM backends, including OpenAI and Hugging Face.

Quickstart

1. Install from PyPI:

   pip install chatways

   or from source:

   git clone https://github.com/zhangsibo1129/chatways.git
   cd chatways
   pip install -e .

2. launch the chat application using chatways command:

   chatways simple --llm-engine fake --port 7860

   • simple starts a basic chat application.
   • --llm-engine fake uses a simulated LLM engine, requiring no local GPU or remote API.
   • --port 7860 sets the port address. For detailed parameter information, refer to the templates.

3. Open your browser and visit http://localhost:7860 to interact with the chat application.

   

How to use

Launch Templates with CLI Command

Chatways offers a range of ready-to-use templates. Templates applications can be quickly launched with a chatways command, and LLM backends can be easily swapped, making it ideal for no-code builders. The command structure is as follows:

chatways [template] [options]

• template: Specifies the type of application (e.g., simple, comparison, arxiv).
• options: Configuration parameters:
  • --llm-engine (-le): LLM engine to use (e.g., openai, huggingface).
  • --llm-model (-lm): LLM model to use (e.g., llama).
  • --llm-config (-lc): Additional configuration in JSON format (e.g., {"device_map":"auto"}).

Here are some examples:

• launch a simple chat application with openai backend

  export OPENAI_API_KEY="openai_api_key"
  export OPENAI_BASE_URL="openai_base_url"
  chatways simple -le openai -lm gpt-3.5-turbo

  or

  chatways simple \
      --llm-engine openai \
      --llm-model gpt-3.5-turbo \
      --llm-model-config '{"api_key":"openai_api_key","base_url":"openai_base_url"}'

• launch a Chat with arXiv application with huggingface backend:

  chatways arxiv \
      --llm-engine huggingface \
      --llm-model Qwen/Qwen1.5-4B-Chat \
      --llm-model-config '{"torch_dtype":"auto","device_map":"auto"}'

Quick Customization with Templates

The src/chatways/template directory contains templates built with the chatways library. Each application template is an independently runnable Gradio app. Users can quickly customize these templates, making it much easier than developing an application from scratch.

Integration with Bot Module

The bot module provides a highly abstract interface designed to simplify integration. This module allows developers to seamlessly integrate LLM into their workflow.

1. Create a chatbot instance

   >>> from chatways import SimpleChatBot
   
   >>> bot = SimpleChatBot(
   ...     llm_engine = "huggingface",
   ...     model = "Qwen/Qwen1.5-4B-Chat",
   ...     model_config = {
   ...         "torch_dtype":"auto",
   ...         "device_map":"auto"
   ...     }
   ... )

2. Chat with the bot

   >>> history = []
   >>> query = "Say three positive words"
   >>> response = bot.chat(query, history=history, stream=False)
   >>> print(response)
   1. hope
   2. confidence 
   3. inspiration

   >>> history.append([query, response]) # Maintain Conversation History
   >>> query = "Please use the second one in a sentence"
   >>> response = bot.chat(query, history=history, stream=False)
   >>> print(response)
   The confidence he had in himself gave him strength to face any challenge that came his way.

Templates

Currently, Chatways provides the following application templates:

Category	Feature	Command
Chat	Simple chat application that enables multi-round conversations with LLM	chatways simple [options]
Chat Comparison	Chat with 2 LLMs simultaneously and compare their results	chatways comparison [options]
Chat with arXiv	Using LLM to make finding academic papers on arXiv more natural and insightful	chatways arxiv [options]

Contributing

As this project is in its early stages, all kinds of contributions are welcome, whether it's reporting issues, suggesting new features, or submitting pull requests.

License

This project is licensed under the MIT License.