Default arguments to use when making requests to the LLM

Usage

chattr_defaults(
  type = "default",
  prompt = NULL,
  max_data_files = NULL,
  max_data_frames = NULL,
  include_doc_contents = NULL,
  include_history = NULL,
  provider = NULL,
  path = NULL,
  model = NULL,
  model_arguments = NULL,
  system_msg = NULL,
  yaml_file = "chattr.yml",
  force = FALSE,
  label = NULL,
  ...
)

Arguments

type: Entry point to interact with the model. Accepted values: 'notebook', chat'
prompt: Request to send to LLM. Defaults to NULL
max_data_files: Sets the maximum number of data files to send to the model. It defaults to 20. To send all, set to NULL
max_data_frames: Sets the maximum number of data frames loaded in the current R session to send to the model. It defaults to 20. To send all, set to NULL
include_doc_contents: Send the current code in the document
include_history: Indicates whether to include the chat history when every time a new prompt is submitted
provider: The name of the provider of the LLM. Today, only "openai" is is available
path: The location of the model. It could be an URL or a file path.
model: The name or path to the model to use.
model_arguments: Additional arguments to pass to the model as part of the request, it requires a list. Examples of arguments: temperature, top_p, max_tokens
system_msg: For OpenAI GPT 3.5 or above, the system message to send as part of the request
yaml_file: The path to a valid config YAML file that contains the defaults to use in a session
force: Re-process the base and any work space level file defaults
label: Label to display in the Shiny app, and other locations
...: Additional model arguments that are not standard for all models/backends

Value

An 'ch_model' object that contains the current defaults that will be used to communicate with the LLM.

Details

The idea is that because we will use addin shortcut to execute the request, all of the other arguments can be controlled via this function. By default, it will try to load defaults from a config YAML file, if none are found, then the defaults for GPT 3.5 will be used. The defaults can be modified by calling this function, even after the interactive session has started.