uqlm.longform.decomposition.response_decomposer.ResponseDecomposer#

class uqlm.longform.decomposition.response_decomposer.ResponseDecomposer(claim_decomposition_llm=None, response_template='zhang_2025')#

Bases: object

__init__(claim_decomposition_llm=None, response_template='zhang_2025')#

Class for decomposing responses into individual claims or sentences. This class is used as an intermediate step for longform UQ methods.

Parameters:

  • claim_decomposition_llm (langchain BaseChatModel, default=None) – A langchain llm BaseChatModel. User is responsible for specifying temperature and other relevant parameters to the constructor of their llm object.

  • response_template (Union[str, Callable], default="zhang_2025") – Specifies the claim decomposition prompt template. Can be one of the following strings: "zhang_2025", "farquhar_2024", "mohri_2024", "jiang_2024", or a custom callable that takes a response string and returns a prompt string.

Methods

__init__([claim_decomposition_llm, ...])

Class for decomposing responses into individual claims or sentences.

decompose_candidate_claims(sampled_responses)

decompose_candidate_sentences(sampled_responses)

decompose_claims(responses[, ...])

decompose_sentences(responses[, progress_bar])

async decompose_candidate_claims(sampled_responses, progress_bar=None)#
Return type:

List[List[List[str]]]

Parameters:

  • sampled_responses (List[List[str]]) – List of lists of sampled responses to be decomposed

  • progress_bar (rich.progress.Progress, default=None) – If provided, displays a progress bar while scoring responses

decompose_candidate_sentences(sampled_responses, progress_bar=None)#
Return type:

List[List[List[str]]]

Parameters:

  • sampled_responses (List[List[str]]) – List of lists of sampled responses to be decomposed

  • progress_bar (rich.progress.Progress, default=None) – If provided, displays a progress bar while scoring responses

async decompose_claims(responses, response_template=None, progress_bar=None)#
Return type:

List[List[str]]

Parameters:

  • responses (List[str]) – LLM response that will be decomposed into independent claims.

  • response_template (Union[str, Callable], default=None) – A string key (e.g. "zhang_2025") or callable that takes a response and returns a prompt. If None, uses the template set at initialization.

  • progress_bar (rich.progress.Progress, default=None) – If provided, displays a progress bar while scoring responses

decompose_sentences(responses, progress_bar=None)#
Return type:

List[List[str]]

Parameters:

  • responses (List[str]) – LLM response that will be decomposed into independent claims.

  • progress_bar (rich.progress.Progress, default=None) – If provided, displays a progress bar while scoring responses

References