Config

Config#

Overview#

In synthora, configurations are defined using YAML files, enabling a modular and reusable approach to setting up models, agents, and tools. Below is a detailed explanation of the configurations and their use cases.

Basic Model Configuration#

The basic_model.yaml file defines the foundational configuration for a model.

name: gpt-4o
backend: openai
backend_config:
  api_key: YOUR_API_KEY
  url: https://api.openai.com
config:
  temperature: 0
  top_p: 1.0

Key Parameters:#

  • name: The name of the model, here gpt-4o.

  • backend: Specifies the backend provider, such as openai.

  • backend_config: Contains the necessary configuration for the backend, like API keys and endpoint URLs.

  • config: Defines model-specific parameters like temperature and top_p for controlling response randomness and diversity.

Referencing a Model Configuration in an Agent#

The basic_agent.yaml file demonstrates how to use the basic_model.yaml configuration in an agent.

name: test
type: vanilla
model: !include basic_model.yaml
prompt: "Your name is {x}"

Key Parameters:#

  • name: The name of the agent, here test.

  • type: The type of the agent, such as vanilla.

  • model: References the model configuration using !include basic_model.yaml.

  • prompt: The agent-specific prompt template, e.g., "Your name is {x}".

Loading the Agent Configuration:#

You can load the agent configuration programmatically:

config = AgentConfig.load("YOUR_PATH/basic_agent.yaml")

Using an Agent as a Tool#

The vanilla_agent.yaml file demonstrates how to embed an agent within another agent or toolkit.

name: Vanilla
type: vanilla
model: !include basic_model.yaml
prompt: "You are an AI Assistant. You are here to help me with my queries. Your name is vanilla."
tools:
  - synthora.toolkits.basic_math_toolkit.BasicMathToolkit
  - target: synthora.agents.VanillaAgent
    args: !include basic_agent.yaml

Key Additions:#

  • tools: Specifies tools or other agents that the current agent can use.

    • synthora.toolkits.basic_math_toolkit.BasicMathToolkit: A toolkit providing basic math functionalities.

    • target: Points to another agent class, here synthora.agents.VanillaAgent.

    • args: Includes configuration for the target agent using !include.

Referencing Predefined Prompts#

The searcher.yaml file illustrates how to use predefined prompts.

name: WebSearcher
description: "This agent is a simple AI assistant that can help you search the web."
type: react
model: !include basic_model.yaml
prompt: !prompt FewShotReactPrompt

Highlights:#

  • prompt: Uses !prompt to reference a predefined prompt (FewShotReactPrompt), allowing for prompt standardization across agents.

Building an Agent from Config#

To construct an agent using its configuration, you can use the appropriate agent class and call the from_config method. The class is determined by the type of the agent.

Example:#

For a VanillaAgent:

agent = VanillaAgent.from_config(config)

This configuration-based setup ensures flexibility, reusability, and a clear separation of concerns, allowing you to define, extend, and manage agents and tools efficiently.

Contents