Welcome to the Ame Module System documentation. This guide covers everything related to modules, including the module handler (module_handler.py), the module API in the controller, and creating custom modules.
To ensure clarity throughout this documentation, let’s define some key terms:
46 |module_handler.py responsible for managing modules.__init__(): model_path: (string), vectorizer_path: (string), modulesjson_path: (string)
55 | Initializes the module handler.
model_path: (string) Path to the module classification model.vectorizer_path: (string) Path to the vectorizer file.modulesjson_path: (string) Path to the modules.json configuration file.load_models(): model_path: (string), vectorizer_path: (string)
63 | Loads the required models for module classification.
model_path: (string) Path to the module classification model.vectorizer_path: (string) Path to the vectorizer file.predict_module(): query: (string)
70 | Predicts a module using the built-in classification model.
query: (string) The user’s query to predict a module.load_modules()
76 | Loads the modules that are marked as detectable.
use_module(): module_name: (string), *args, **kwargs
79 | Uses a module and obtains a response from its associated function.
module_name: (string) Name of the module to use.*args: Variable-length positional arguments.**kwargs: Variable-length keyword arguments.To run a module using the API, follow these steps:
87 |from controller import Controller
88 |
89 | # Initialize the controller
90 | controller = Controller()
91 |
92 | # Run a module
93 | response = controller.run_module(module, args)
94 | To create custom modules, modify the modules.json file with the following keys:
id: (integer)name: (string)args: (union[None, dict])None, the module is detected using predict_module() and handled as a single-pass module. If dict, the module is detected, and its arguments are passed to an LLM to parse the query (double-pass module).function: (string)detectable: (boolean)True, the module is detectable via the built-in classifier.Place your module file (.py) in the modules folder.
The classifier is used for both single-pass and double-pass modules. To train your module to be detected by the built-in classifier:
112 |Modify the training data (module_engine/training_data.csv) and validation data (module_engine/validation_data.csv) by adding pairs of queries and labels in CSV format.
Example:
118 |query,label
119 | Tell me the latest finance news,News
120 | What's the current dew point?,Weather
121 | Use a Language Model (LLM) to generate these pairs, such as ChatGPT, or handwrite them.
125 |Ensure that the training data and validation data are unique.
128 |Start the training by running:
131 |python module_engine/training/train.py
132 | Thank you for exploring the Ame Module System documentation. This system empowers you to create and manage custom modules seamlessly. If you have any questions or need further assistance, please don’t hesitate to reach out.
137 |For additional resources, including more documentation and support forums, visit the discord server.
138 | 139 |