vertebrae.utils.embedding_batches

Deterministic batching helpers shared by endpoint embedding workflows.

Functions

endpoint_n_rows(values)	Return aligned endpoint row count for common input containers.
take_endpoint_rows(values, indices)	Select endpoint rows while preserving common container types.
iter_endpoint_batches(values, batch_size)	Yield deterministic contiguous endpoint batches.
encode_endpoint_batches(values, *, batch_size, encode, ...)	Encode and combine deterministic endpoint batches without sparse densification.

Module Contents

vertebrae.utils.embedding_batches.endpoint_n_rows(values)

Return aligned endpoint row count for common input containers.

Parameters:

values (Any)

Return type:

int

vertebrae.utils.embedding_batches.take_endpoint_rows(values, indices)

Select endpoint rows while preserving common container types.

Parameters:

• values (Any)

• indices (numpy.ndarray)

Return type:

Any

vertebrae.utils.embedding_batches.iter_endpoint_batches(values, batch_size)

Yield deterministic contiguous endpoint batches.

Parameters:

• values (Any)

• batch_size (int)

Return type:

Iterator[Tuple[numpy.ndarray, Any]]

vertebrae.utils.embedding_batches.encode_endpoint_batches(values, *, batch_size, encode, owner, profiler=None, call_type='encode_endpoint', memory_config=None)

Encode and combine deterministic endpoint batches without sparse densification.

Encoded rows and their final ordering references are admitted progressively through shared matrix and metadata stagers. No-spill runs fail as soon as retaining the next row or reference would cross the configured budget; spill-enabled runs write both to temporary storage immediately. Omitting memory_config uses the default memory policy.

Parameters:

• values (Any)

• batch_size (int)

• encode (Callable[[Any], Any])

• owner (str)

• profiler (Any)

• call_type (str)

• memory_config (Optional[vertebrae.config.MemoryConfig])

Return type:

Any