Skip to content

Latest commit

 

History

History
308 lines (127 loc) · 7.8 KB

nemoguardrails.rails.llm.config.md

File metadata and controls

308 lines (127 loc) · 7.8 KB

module nemoguardrails.rails.llm.config

Module for the configuration of rails.

class Model

Configuration of a model used by the rails engine.

Typically, the main model is configured e.g.: { "type": "main", "engine": "openai", "model": "gpt-3.5-turbo-instruct" }

class Instruction

Configuration for instructions in natural language that should be passed to the LLM.

class Document

Configuration for documents that should be used for question answering.

class SensitiveDataDetectionOptions

class SensitiveDataDetection

Configuration of what sensitive data should be detected.

class MessageTemplate

Template for a message structure.

class TaskPrompt

Configuration for prompts that will be used for a specific task.

classmethod TaskPrompt.check_fields

check_fields(values)

class EmbeddingSearchProvider

Configuration of a embedding search provider.

class KnowledgeBaseConfig

class CoreConfig

Settings for core internal mechanics.

class InputRails

Configuration of input rails.

class OutputRails

Configuration of output rails.

class RetrievalRails

Configuration of retrieval rails.

class SingleCallConfig

Configuration for the single LLM call option for topical rails.

class UserMessagesConfig

Configuration for how the user messages are interpreted.

class DialogRails

Configuration of topical rails.

class FactCheckingRailConfig

Configuration data for the fact-checking rail.

class RailsConfigData

Configuration data for specific rails that are supported out-of-the-box.

class Rails

Configuration of specific rails.

class RailsConfig

Configuration object for the models and the rails.

TODO: add typed config for user_messages, bot_messages, and flows.

property RailsConfig.streaming_supported

Whether the current config supports streaming or not.

Currently, we don't support streaming if there are output rails.

method RailsConfig.from_content

from_content(
    colang_content: Optional[str] = None,
    yaml_content: Optional[str] = None,
    config: Optional[dict] = None
)

Loads a configuration from the provided colang/YAML content/config dict.

method RailsConfig.from_path

from_path(
    config_path: str,
    test_set_percentage: Optional[float] = 0.0,
    test_set: Optional[Dict[str, List]] = {},
    max_samples_per_intent: Optional[int] = 0
)

Loads a configuration from a given path.

Supports loading a from a single file, or from a directory.

Also used for testing Guardrails apps, in which case the test_set is randomly created from the intent samples in the config files. In this situation test_set_percentage should be larger than 0.

If we want to limit the number of samples for an intent, set the max_samples_per_intent to a positive number. It is useful for testing apps, but also for limiting the number of samples for an intent in some scenarios. The chosen samples are selected randomly for each intent.

classmethod RailsConfig.parse_object

parse_object(obj)

Parses a configuration object from a given dictionary.