Skip to main content

agrag.chunking.turns.TurnWindowChunker

Bases: Chunker

Packs whole chat turns into windows of at most chunk_size tokens.

The chunker reads Document.turns. A window holds one or more whole turns, and a chunk boundary never falls inside a turn. Tokens are counted for each turn alone, so the separators between turns are not part of the count. Text before the first turn joins the first window, and text after a turn joins the window that holds that turn.

A turn above the budget is split by fallback, and its chunks have the chunker name turn-window:<fallback strategy>. A document without turns is split by fallback as a whole and its chunks have the same name.

Attributes:

  • chunk_size (int) – The most tokens in a window, counted with tokenizer.
  • turn_overlap (int) – The number of turns that a window repeats from the window before it. A window always moves on by at least one turn.
  • tokenizer (str) – The tokenizer that counts size. "character" counts characters.
  • fallback (SerializeAsAny[SpanChunker]) – The chunker for a turn above the budget and for a document without turns.

Functions:

  • chunk – Split a document into chunks.
  • fingerprint – Return the hash of settings(), 16 hex characters.
  • model_copy – Copy the chunker, validating any changed setting.
  • model_post_init – Load the tokenizer once, so a bad name fails at construction.
  • settings – Return the strategy name and every setting as JSON-safe data.

chunk​

chunk(document:Document) -> list[Chunk]

Split a document into chunks.

Every chunk has non-empty text, indexes run from 0 without gaps, and a chunk with text provenance has text equal to document.text at its offsets.

Parameters:

  • document (Document) – The document to split.

Returns:

  • list[Chunk] – The chunks, in document order, each with chunker and chunker_hash set. A strategy that sets chunker itself keeps its value.

Raises:

  • ChunkingError – The strategy returned chunks that break the contract.

chunk_size​

chunk_size: int = Field(default=256, gt=0)

fallback​

fallback: SerializeAsAny[SpanChunker] = Field(default_factory=RecursiveChunker)

fingerprint​

fingerprint() -> str

Return the hash of settings(), 16 hex characters.

model_config​

model_config = ConfigDict(frozen=True, extra='forbid')

model_copy​

model_copy(*, update:Mapping[str, Any] | None = None, deep:bool = False) -> Self

Copy the chunker, validating any changed setting.

A plain copy keeps the fingerprint and the splitter of the original. A copy with changes is built again from its configuration.

model_post_init​

model_post_init(context:Any) -> None

Load the tokenizer once, so a bad name fails at construction.

settings​

settings() -> dict[str, Any]

Return the strategy name and every setting as JSON-safe data.

A setting that is itself a chunker appears as that chunker's settings.

strategy​

strategy: str

The strategy name, "turn-window".

tokenizer​

tokenizer: str = DEFAULT_TOKENIZER

turn_overlap​

turn_overlap: int = Field(default=0, ge=0)