Struct RefineDocumentsChain 

pub struct RefineDocumentsChain { /* private fields */ }

RefineDocumentsChain

Iteratively refines the answer document by document. Generates an initial answer from the first document, then refines with each subsequent document.

Implementations

impl RefineDocumentsChain

pub fn new<L>(llm: L) -> Self

where L: BaseChatModel + Send + Sync + 'static, L::Error: Into<ProviderError>,

Create a new RefineDocumentsChain with the given LLM.

pub fn with_initial_prompt(self, template: impl Into<String>) -> Self

Set the initial prompt template used for the first document.

pub fn with_refine_prompt(self, template: impl Into<String>) -> Self

Set the refine prompt template used for subsequent documents.

pub fn with_document_variable(self, name: impl Into<String>) -> Self

Set the document variable name used in the prompts.

pub fn with_input_key(self, key: impl Into<String>) -> Self

Set the input key.

pub fn with_output_key(self, key: impl Into<String>) -> Self

Set the output key.

pub fn with_name(self, name: impl Into<String>) -> Self

Set the chain name.

pub fn with_verbose(self, verbose: bool) -> Self

Set verbose mode.

pub fn build_initial_prompt(&self, context: &str, input: &str) -> String

Build the initial prompt from the first document’s content and input.

pub fn build_refine_prompt( &self, context: &str, input: &str, existing_answer: &str, ) -> String

Build the refine prompt from the new context, input, and existing answer.

pub async fn invoke_with_documents( &self, documents: Vec<Document>, input: &str, ) -> Result<String, ChainError>

Invoke with documents and input directly (iterative refinement).

Trait Implementations

impl BaseChain for RefineDocumentsChain

fn stream<'life0, 'async_trait>( &'life0 self, inputs: HashMap<String, Value>, ) -> Pin<Box<dyn Future<Output = Result<ChainStream, ChainError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

Stream execution for RefineDocumentsChain.

Runs the initial + all intermediate refine steps via invoke (since their output feeds the next step), then streams the final refine step token by token via stream_chat. With a single document there is no final refine step — the initial answer is emitted directly (P2-4), so the LLM is not re-called on the identical initial prompt.

fn input_keys(&self) -> Vec<&str>

Get input keys.

fn output_keys(&self) -> Vec<&str>

Get output keys.

fn invoke<'life0, 'async_trait>( &'life0 self, inputs: HashMap<String, Value>, ) -> Pin<Box<dyn Future<Output = Result<ChainResult, ChainError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

Execute the Chain.

fn name(&self) -> &str

Get Chain name.

fn invoke_with_config<'life0, 'async_trait>( &'life0 self, inputs: HashMap<String, Value>, config: Option<RunnableConfig>, ) -> Pin<Box<dyn Future<Output = Result<ChainResult, ChainError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

Execute the Chain with a RunnableConfig.

fn stream_with_config<'life0, 'async_trait>( &'life0 self, inputs: HashMap<String, Value>, config: Option<RunnableConfig>, ) -> Pin<Box<dyn Future<Output = Result<ChainStream, ChainError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

Stream execute the Chain with a RunnableConfig.

fn validate_inputs( &self, inputs: &HashMap<String, Value>, ) -> Result<(), ChainError>

Validate inputs.