Reference

class tca.pipeline.SearchIndex(embeddings, metadata, config)

Parameters:

• embeddings (np.ndarray)

• metadata (list[MetadataRecord])

• config (TurboQuantConfig)

classmethod from_embeddings(embeddings, metadata, config=None)

Parameters:

• embeddings (ndarray)

• metadata (list[dict[str, Any]])

• config (TurboQuantConfig | None)

Return type:

SearchIndex

search(query, filters=None, *, query_mode='auto')

Parameters:

• query (ndarray)

• filters (dict[str, Any] | None)

• query_mode (str)

Return type:

SearchResults

search_exact(query, filters=None)

Parameters:

• query (ndarray)

• filters (dict[str, Any] | None)

Return type:

SearchResults

to_manifest()

Return type:

dict[str, Any]

class tca.config.TurboQuantConfig(bit_width: 'int' = 3, candidate_k: 'int' = 128, rerank_k: 'int' = 20, oversample: 'int' = 2, seed: 'int' = 0, quantizer_kind: 'str' = 'prod', lloyd_max_iter: 'int' = 100, lloyd_tol: 'float' = 1e-06, monte_carlo_samples: 'int' = 20000, store_original_embeddings: 'bool' = True, auto_score_gap_threshold: 'float' = 0.06, auto_score_spread_threshold: 'float' = 0.015, max_candidate_k: 'int' = 2048, max_oversample: 'int' = 8)

Parameters:

• bit_width (int)

• candidate_k (int)

• rerank_k (int)

• oversample (int)

• seed (int)

• quantizer_kind (str)

• lloyd_max_iter (int)

• lloyd_tol (float)

• monte_carlo_samples (int)

• store_original_embeddings (bool)

• auto_score_gap_threshold (float)

• auto_score_spread_threshold (float)

• max_candidate_k (int)

• max_oversample (int)

bit_width: int

candidate_k: int

rerank_k: int

oversample: int

seed: int

quantizer_kind: str

lloyd_max_iter: int

lloyd_tol: float

monte_carlo_samples: int

store_original_embeddings: bool

auto_score_gap_threshold: float

auto_score_spread_threshold: float

max_candidate_k: int

max_oversample: int

validate()

Return type:

TurboQuantConfig

to_dict()

Return type:

dict[str, Any]

to_json(path)

Parameters:

path (str | Path)

Return type:

None

classmethod from_json(path)

Parameters:

path (str | Path)

Return type:

TurboQuantConfig

tca.quantization.fit_scalar_codebook(bit_width, dimension, n_samples=20000, seed=0, max_iter=100, tol=1e-06)

Parameters:

• bit_width (int)

• dimension (int)

• n_samples (int)

• seed (int)

• max_iter (int)

• tol (float)

Return type:

ndarray[tuple[Any, …], dtype[float32]]

class tca.quantization.EncodedMSE(indices: 'IntArray')

Parameters:

indices (ndarray[tuple[Any, ...], dtype[int32]])

indices: ndarray[tuple[Any, ...], dtype[int32]]

class tca.quantization.TurboQuantMSE(dimension, bit_width, *, seed=0, monte_carlo_samples=20000, lloyd_max_iter=100, lloyd_tol=1e-06)

Parameters:

• dimension (int)

• bit_width (int)

• seed (int)

• monte_carlo_samples (int)

• lloyd_max_iter (int)

• lloyd_tol (float)

encode(x)

Parameters:

x (_Buffer | _SupportsArray[dtype[Any]] | _NestedSequence[_SupportsArray[dtype[Any]]] | complex | bytes | str | _NestedSequence[complex | bytes | str])

Return type:

EncodedMSE

decode(encoded)

Parameters:

encoded (EncodedMSE | _Buffer | _SupportsArray[dtype[Any]] | _NestedSequence[_SupportsArray[dtype[Any]]] | complex | bytes | str | _NestedSequence[complex | bytes | str])

Return type:

ndarray[tuple[Any, …], dtype[float32]]

prepare_query(query)

Parameters:

query (_Buffer | _SupportsArray[dtype[Any]] | _NestedSequence[_SupportsArray[dtype[Any]]] | complex | bytes | str | _NestedSequence[complex | bytes | str])

Return type:

ndarray[tuple[Any, …], dtype[float32]]

approximate_inner_products(query, encoded)

Parameters:

• query (_Buffer | _SupportsArray[dtype[Any]] | _NestedSequence[_SupportsArray[dtype[Any]]] | complex | bytes | str | _NestedSequence[complex | bytes | str])

• encoded (EncodedMSE | _Buffer | _SupportsArray[dtype[Any]] | _NestedSequence[_SupportsArray[dtype[Any]]] | complex | bytes | str | _NestedSequence[complex | bytes | str])

Return type:

ndarray[tuple[Any, …], dtype[float32]]

class tca.quantization.EncodedProd(indices: 'IntArray', signs: 'SignArray', residual_norms: 'FloatArray')

Parameters:

• indices (ndarray[tuple[Any, ...], dtype[int32]])

• signs (ndarray[tuple[Any, ...], dtype[int8]])

• residual_norms (ndarray[tuple[Any, ...], dtype[float32]])

indices: ndarray[tuple[Any, ...], dtype[int32]]

signs: ndarray[tuple[Any, ...], dtype[int8]]

residual_norms: ndarray[tuple[Any, ...], dtype[float32]]

class tca.quantization.TurboQuantProd(dimension, bit_width, *, seed=0, monte_carlo_samples=20000, lloyd_max_iter=100, lloyd_tol=1e-06)

Parameters:

• dimension (int)

• bit_width (int)

• seed (int)

• monte_carlo_samples (int)

• lloyd_max_iter (int)

• lloyd_tol (float)

encode(x)

Parameters:

x (_Buffer | _SupportsArray[dtype[Any]] | _NestedSequence[_SupportsArray[dtype[Any]]] | complex | bytes | str | _NestedSequence[complex | bytes | str])

Return type:

EncodedProd

decode(encoded)

Parameters:

encoded (EncodedProd)

Return type:

ndarray[tuple[Any, …], dtype[float32]]

prepare_query(query)

Parameters:

query (_Buffer | _SupportsArray[dtype[Any]] | _NestedSequence[_SupportsArray[dtype[Any]]] | complex | bytes | str | _NestedSequence[complex | bytes | str])

Return type:

tuple[ndarray[tuple[Any, …], dtype[float32]], ndarray[tuple[Any, …], dtype[float32]]]

approximate_inner_products(query, encoded)

Parameters:

• query (_Buffer | _SupportsArray[dtype[Any]] | _NestedSequence[_SupportsArray[dtype[Any]]] | complex | bytes | str | _NestedSequence[complex | bytes | str])

• encoded (EncodedProd)

Return type:

ndarray[tuple[Any, …], dtype[float32]]

tca.quantization.exact_topk(query, bank, top_k, ids=None)

Parameters:

• query (_Buffer | _SupportsArray[dtype[Any]] | _NestedSequence[_SupportsArray[dtype[Any]]] | complex | bytes | str | _NestedSequence[complex | bytes | str])

• bank (_Buffer | _SupportsArray[dtype[Any]] | _NestedSequence[_SupportsArray[dtype[Any]]] | complex | bytes | str | _NestedSequence[complex | bytes | str])

• top_k (int)

• ids (Iterable[int] | None)

Return type:

tuple[ndarray, ndarray]

tca.cli.build_parser()

Return type:

ArgumentParser

tca.cli.main()

Return type:

None