usersimcrs.nlg.llm.nlg_generative_llm

Generative natural language generation using a large language model.

Note that the large language model is expected to be hosted externally. Also, the prompt used to generate utterances is expected to have the following placeholders: - “dialogue_acts”: to be replaced by the stringified dialogue acts. - “annotations”: to be replaced by the stringified annotations (if provided).

Classes

LLMGenerativeNLG

Initializes the generative NLG.

Module Contents

class usersimcrs.nlg.llm.nlg_generative_llm.LLMGenerativeNLG(llm_interface: usersimcrs.llm_interfaces.llm_interface.LLMInterface, prompt_file: str, prompt_prefix: str | None = None)

Bases: dialoguekit.nlg.nlg_abstract.AbstractNLG

Initializes the generative NLG.

Parameters:

llm_interface – Interface to the large language model.
prompt_file – Path to the prompt file.
prompt_prefix – Prefix to be remove from generated utterances. Defaults to None.

Raises:

FileNotFoundError – If the prompt file is not found.

llm_interface

prompt

prompt_prefix = None

generate_utterance_text(dialogue_acts: List[dialoguekit.core.dialogue_act.DialogueAct], annotations: List[dialoguekit.core.annotation.Annotation] | None | None = None, force_annotation: bool = False) → dialoguekit.core.annotated_utterance.AnnotatedUtterance | bool

Turns a structured utterance into a textual one.

Parameters:

dialogue_acts – Dialogue acts of the utterance to be generated.
annotations – If provided, these annotations should be considered during generation.
force_annotation – A flag to indicate whether annotations should be forced or not. Not used in this NLG.

Raises:

RuntimeError – If generation fails.

Returns:

Natural language utterance.