hbllmutils.model.load

Utilities for loading Large Language Model (LLM) configurations and creating model instances.

This module focuses on creating RemoteLLMModel instances from configuration files/directories or explicit runtime parameters. It also provides a convenient dispatcher for loading a model by name, reusing an existing instance, or falling back to the default configuration.

The module contains the following main components:

Note

Configuration loading is delegated to hbllmutils.manage.LLMConfig. If configuration is missing and no explicit parameters are provided, a RuntimeError is raised.

Example:

>>> from hbllmutils.model.load import load_llm_model, load_llm_model_from_config
>>> # Load a configured model by name (from config files)
>>> model = load_llm_model("gpt-4")
>>> # Load the default model (from config files)
>>> default_model = load_llm_model()
>>> # Create a model directly with explicit parameters
>>> direct_model = load_llm_model_from_config(
...     base_url="https://api.example.com/v1",
...     api_token="your-token",
...     model_name="gpt-4"
... )

LLMModelTyping

hbllmutils.model.load.LLMModelTyping

Type alias for model input, which can be either a string (model name) or an LLMModel instance.

alias of str | LLMModel

load_llm_model_from_config

hbllmutils.model.load.load_llm_model_from_config(config_file_or_dir: str | None = None, base_url: str | None = None, api_token: str | None = None, model_name: str | None = None, **params: Any) → RemoteLLMModel[source]

Load a Large Language Model with specified configuration.

This function attempts to load LLM configuration from a config file or directory, and creates a remote model instance. It supports both pre-configured models from config files and dynamically specified configurations.

Parameters:

  • config_file_or_dir (Optional[str]) – Path to the configuration file or directory. If None, defaults to the current directory.

  • base_url (Optional[str]) – Base URL for the LLM API endpoint. If provided, overrides config file settings.

  • api_token (Optional[str]) – API token for authentication. Required when base_url is provided without a config file.

  • model_name (Optional[str]) – Name of the model to load. Required when base_url is provided without a config file.

  • params (Any) – Additional parameters to pass to the model.

Returns:

An initialized LLM remote model instance.

Return type:

RemoteLLMModel

Raises:

  • FileNotFoundError – When config file is not found (handled internally).

  • KeyError – When specified model is not found in config (handled internally).

  • ValueError – When api_token is not specified but required, or when model_name is empty but required.

  • RuntimeError – When no model parameters are specified and no local configuration is available.

Example::
>>> # Load model from config file
>>> model = load_llm_model_from_config(config_file_or_dir='./config')
>>>
>>> # Load model with explicit parameters
>>> model = load_llm_model_from_config(
...     base_url='https://api.example.com',
...     api_token='your-token',
...     model_name='gpt-4'
... )
>>>
>>> # Load model from config with overrides
>>> model = load_llm_model_from_config(
...     config_file_or_dir='./config',
...     model_name='gpt-4',
...     base_url='https://custom-api.example.com'
... )

load_llm_model

hbllmutils.model.load.load_llm_model(model: str | LLMModel | None = None) → LLMModel[source]

Load a Large Language Model from various input types.

This is a convenience function that handles different types of model specifications. It can load a model by name from configuration, use an existing model instance, or load the default model from configuration.

Parameters:

model (Optional[LLMModelTyping]) – The model specification. Can be: - A string representing the model name to load from configuration - An LLMModel instance to use directly - None to load the default model from configuration

Returns:

An initialized LLM model instance.

Return type:

LLMModel

Raises:

  • TypeError – When model is not a string, LLMModel instance, or None.

  • ValueError – When model name is invalid or not found in configuration.

  • RuntimeError – When no model parameters are specified and no local configuration is available.

Example::
>>> # Load model by name from configuration
>>> model = load_llm_model('gpt-4')
>>>
>>> # Use an existing model instance
>>> existing_model = RemoteLLMModel(base_url='...', api_token='...', model_name='gpt-4')
>>> model = load_llm_model(existing_model)
>>>
>>> # Load default model from configuration
>>> model = load_llm_model()