Skip to main content
Pre-release

This release is a pre-release and may not be stable for production use.

QaNN

QaNN (Quantized Nearest Neighbors) is a Python library for fast nearest-neighbor search over dense float32 vectors and sparse vectors. For dense vectors it offers exact search, inverted-file (IVF) and HNSW graph search, product and scalar quantization and exact re-ranking; for sparse vectors (BM25, SPLADE), exact and approximate inverted-index search. Every index supports custom ids, deletion, metadata filters and saving. All of it is implemented in multithreaded, SIMD-accelerated C++ and driven from NumPy and SciPy.

Alpha: under active development. The API may change between releases.

Install

pip install qann

Wheels are published for Linux (x86_64, aarch64) and macOS (Apple Silicon, Intel) on Python 3.10 to 3.14. x86_64 builds require AVX2 and FMA. The sparse indexes take scipy.sparse matrices, so they also need SciPy (pip install scipy). To build from source instead, run pip install . in a checkout; this needs CMake 3.20+, a C++20 compiler and, on Linux, OpenBLAS (optional but much faster training).

Quick start

import numpy as np, qann

x = np.random.rand(100_000, 128).astype(np.float32)

ivf = qann.IVFIndex(128, nlist=1024, nprobe=16)
ivf.enable_pq(16)                           # store each vector in 16 bytes
ivf.train(x, seed=1)
index = qann.RefineIndex(ivf, k_factor=10)  # re-rank PQ candidates exactly
index.add(x)

ids, dists = index.batch_query(x[:5], 10)   # both shaped (5, 10)

Conventions

  • Vectors are passed as 2D NumPy arrays of shape (n, dim), except single vectors (query, PQCodebook methods), which are (dim,). float32 C-contiguous arrays are used without copying; anything else (float64, Fortran order, strided slices) is converted first.
  • IDs are assigned in insertion order by default: the first vector added gets id 0, the next id 1, and so on across all add calls. Create an index with custom_ids=True to use your own ids instead; see Custom ids and deleting.
  • Distances are squared Euclidean (L2) distances by default, cosine distances with metric=qann.Metric.Cosine, or negative inner products with metric=qann.Metric.InnerProduct (see Metrics); smallest first in every case.
  • Results from batch_query are an int64 id array and a float32 distance array, both shaped (num_queries, k). Rows with fewer than k hits are padded with id -1 and distance inf.

Indexes

Indexes come in two families, chosen by the kind of vectors they hold, with one class per search algorithm:

Dense vectors (qann.DenseIndex) Sparse vectors (qann.SparseIndex)
Exact FlatIndex InvertedIndex
Approximate IVFIndex, HNSWIndex, RefineIndex (exact re-ranking of either) SparseApproxIndex

The algorithm is fixed when an index is created; search settings such as nprobe or ef_search are properties that can change at any time. Everything except add, query, batch_query, dim and metric is shared by both families through qann.IndexBase: custom ids, deletion, metadata and filters, saving. qann.Index is the old name of qann.DenseIndex and still works.

Every dense index has these methods (sparse indexes have the same ones, minus dim, taking scipy.sparse matrices; see InvertedIndex):

Method Description
add(data, ids=None, *, metadata=None) Add an (n, dim) array of vectors, with n ids if the index uses custom ids, and optional metadata columns; see Metadata. Returns the vectors' ids.
batch_query(queries, k, *, allowed=None, where=None) k nearest neighbors for each row of an (nq, dim) array, run in parallel. Returns (ids, dists). allowed restricts results to those ids and where to vectors whose metadata matches; see Filtered search.
query(query, k, *, allowed=None, where=None) k nearest neighbors for a single (dim,) vector such as x[i]. Returns 1D (ids, dists).
metadata_schema Each metadata column's name and type ('int', 'float' or 'str').
get_metadata(ids, columns=None) Read metadata back as {name: one value per id}; see Metadata.
update_metadata(ids, metadata) Overwrite metadata columns of existing vectors in place; see Metadata.
add_metadata_column(name, type, *, default=None, ids=None, values=None) Add a metadata column ('int', 'float' or 'str') to the vectors already in the index; see Metadata.
drop_metadata_column(name) Remove a metadata column and its values.
remove(ids) Delete vectors by id; returns how many were deleted.
compact() Free the memory of deleted vectors; returns how many were freed. Ids are unchanged, and so are results, except that HNSWIndex rebuilds its graph.
size() Number of vectors in the index (deleted ones excluded).
dim() Vector dimension.
save(path) Save the index to a file; see Saving and loading.

FlatIndex(dim, *, custom_ids=False)

Exact brute-force search: no training, 100% recall, and query cost that grows linearly with the number of vectors. Good for up to a few hundred thousand vectors, or as ground truth when measuring recall.

index = qann.FlatIndex(128)
index.add(x)
ids, dists = index.batch_query(x[:10], 10)

IVFIndex(dim, nlist, nprobe=10, *, custom_ids=False)

Partitions vectors into nlist clusters with k-means and, at query time, scans only the nprobe clusters closest to the query. Much faster than flat search at a small cost in recall.

Member Description
train(data, max_iters=25, seed=None) Learn the cluster centroids from a sample of at least nlist vectors. Call once, before add. Pass seed for reproducible results.
enable_pq(num_subspaces, centroids_per_subspace=None, *, fast_scan=False) Compress stored vectors with product quantization. Call before train. centroids_per_subspace defaults to 256 (one byte per slice), or 16 with fast_scan; see Fast-scan.
fast_scan Read-only property: whether enable_pq(..., fast_scan=True) was used.
enable_sq(bits=8) Store each vector as scalar codes of 8 or 4 bits per dimension instead of floats (see SQ). Call before train; not together with enable_pq.
sq_bits Read-only property: 8 or 4 after enable_sq(), else 0.
nprobe Property: clusters scanned per query. Can be changed at any time; higher means better recall and slower queries.
pq_distance Property: how PQ codes are scored, qann.PQDistance.ADC (default, more accurate) or qann.PQDistance.SDC (not with Metric.InnerProduct).
precomputed_tables Property: whether ADC queries use lookup tables precomputed at train time, which makes PQ queries faster. On by default; reads True once the tables exist (PQ enabled, trained, and nlist * num_subspaces * centroids_per_subspace * 4 bytes within 256 MB). Set to False to free them. Always False with Metric.InnerProduct, which doesn't need them.

nlist must be between 100 and 65535, and nprobe between 1 and nlist. A common starting point is nlist around sqrt(n) to 4 * sqrt(n) and nprobe at 1% to 5% of nlist.

With PQ (enable_pq), each vector is split into num_subspaces slices and each slice is stored as one byte, so a 128-dim float32 vector (512 bytes) with 16 subspaces takes 16 bytes. dim must be divisible by num_subspaces. Distances become approximate, which lowers recall; wrap the index in a RefineIndex to recover it.

ivf = qann.IVFIndex(128, nlist=1024, nprobe=16)
ivf.enable_pq(16)
ivf.train(x, seed=1)
ivf.add(x)
ivf.nprobe = 32   # trade speed for recall

Fast-scan

enable_pq(num_subspaces, fast_scan=True) stores each slice in 4 bits (16 centroids per slice) and scores 32 vectors at a time with a single SIMD table lookup per slice (NEON on ARM, SSSE3 on x86). Scanning the codes is about 6x faster than regular PQ at the same bytes per vector. Use twice the num_subspaces you would with regular PQ to keep the memory and most of the accuracy, and pair it with a RefineIndex:

ivf = qann.IVFIndex(128, nlist=1024, nprobe=16)
ivf.enable_pq(64, fast_scan=True)          # 64 x 4 bits = 32 bytes per vector
ivf.train(x, seed=1)
refine = qann.RefineIndex(ivf, k_factor=16)
refine.add(x)
  • On SIFT1M with re-ranking (k_factor=16), 64 subspaces with fast-scan match 32 regular subspaces (recall@10 0.931 at nprobe=16, 0.995 at nprobe=64) and beat 16 regular ones, while scanning several times faster.
  • Without a RefineIndex, recall is lower than regular PQ at the same memory, and the returned distances are approximate: the lookup tables are rounded to 6 bits.
  • num_subspaces must be even and at most 1024, centroids_per_subspace must be 16 (the default with fast_scan), and pq_distance stays ADC: setting SDC raises ValueError.

Rotation

enable_pq(..., rotate=True) makes train() also learn a rotation of the vectors (PCA, with the variance spread evenly over the subspaces), and every vector is rotated before IVF and PQ see it. Rotations don't change distances, so results stay valid; only PQ's error changes. It helps when dimensions are correlated across subspaces, so plain PQ spends bits encoding the same information twice:

ivf.enable_pq(64, fast_scan=True, rotate=True)
  • On SIFT1M with fast-scan, recall@10 at nprobe=16 goes from 0.635 to 0.718 without re-ranking, and k_factor=4 reaches what k_factor=8 did (0.988 vs 0.956 at nprobe=64). With re-ranking that is about 5% more queries per second at the same recall, after the cost of rotating each query.
  • It does nothing for embeddings whose dimensions are already uncorrelated (GloVe: within 0.004 either way), and with heavy re-ranking (k_factor=16) there is little left to gain. Measure on your data.
  • Training takes a little longer (0.4 s more on SIFT1M), and the index stores one dim x dim matrix (64 KB at 128 dimensions).

SQ

enable_sq(8) stores each vector as one byte per dimension (on SIFT, 128 bytes instead of 512) and enable_sq(4) as half a byte; train also learns each dimension's range. Unlike PQ there are no codebooks, and 8 bits loses little recall, so it needs no RefineIndex: it is the IVF option for high recall without keeping the raw vectors in memory. 4 bits loses more and should be wrapped in a RefineIndex(k_factor=8).

ivf = qann.IVFIndex(128, nlist=1024, nprobe=16)
ivf.enable_sq(8)
ivf.train(x, seed=1)
ivf.add(x)

On SIFT1M (all cores, busy machine), 8 bits reached recall@10 0.925 at nprobe=16 and 0.985 at 64 (floats: 0.932 and 0.995) at 1.3 to 1.4x the queries per second of float lists, in a quarter of the memory; 4 bits + RefineIndex(k_factor=8) matched the floats' recall exactly at the same speed-up. On one thread the gain mostly disappears: SQ helps because the float scan is limited by memory bandwidth. For the fastest queries when raw vectors fit in memory, fast-scan PQ + RefineIndex is still quicker, and HNSW-SQ8 is faster at the same recall for about twice the memory.

HNSWIndex(dim, M=16, ef_construction=200, *, custom_ids=False, seed=None)

A graph index (Hierarchical Navigable Small World): every vector is linked to M near neighbors (2 * M on the bottom level), and a query walks the graph toward the query vector. No training; add builds the graph on all cores. It stores vectors in full, so distances are exact, and it needs no representative training sample, at the cost of more memory per vector (the vector plus its links) and slower adds. With enable_sq() it stores 8- or 4-bit codes per dimension instead (see below).

Member Description
ef_search Property: how many candidates a query keeps while searching (default 64; queries use at least k). Can be changed at any time; higher means better recall and slower queries.
M, ef_construction Read-only properties: the settings the graph was built with. Higher M or ef_construction builds a better graph, more slowly.
enable_sq(bits=8) Store vectors as scalar codes of bits per dimension: 8 or 4 (anything else raises ValueError). Call before train() and add().
train(x) With SQ: learn each dimension's range from sample vectors. Required before add(); raises RuntimeError without SQ.
sq_bits Read-only property: 8 or 4 after enable_sq(), else 0.
index = qann.HNSWIndex(128, M=16, ef_construction=200)
index.add(x)
index.ef_search = 128   # trade speed for recall
ids, dists = index.batch_query(q, 10)

On SIFT1M (1M vectors), M=16, ef_construction=200 reaches recall@10 0.960 at ef_search=64 (about 80 µs per query on one thread) and 0.995 at 256 (about 340 µs). On all cores, with the machine busy, the float graph built in 75 s and reached 0.996 at 11.4k queries/s; with 8-bit SQ and re-ranking (below) it reached 0.996 at 15.8k/s, close to IVF + fast-scan + RefineIndex at 0.995 and 16.9k/s, which built in 5 s. Below 0.96 recall HNSW is the faster of the two on SIFT1M. On GloVe-100 word embeddings (cosine) the order flips: IVF + fast-scan + RefineIndex was about twice as fast up to 0.94 recall, and HNSW with M=16 needed ef_search=1024 to reach 0.95. Benchmark on your own data before choosing. Choose IVF + fast-scan when you can train on representative data and build time matters; choose HNSW when you can't train up front, or when vectors keep arriving that the IVF clusters were not trained on.

  • Deleting marks vectors deleted; queries still route through them, so the graph stays connected, but never return them. compact() rebuilds the graph from the remaining vectors, which takes about as long as building it again, so call it after many deletes rather than after each one.
  • Builds are parallel, so the graph differs slightly from run to run. With seed and qann.set_num_threads(1), the build is reproducible.
  • Choosing M: the default 16 suits data like SIFT. For high-dimensional embeddings (word or sentence vectors), use M=32 to 48 with ef_construction=300 to 400: on GloVe-100, M=48 answered 1.5x more queries per second than M=16 at 0.90 and 0.95 recall, and only M=48 reached 0.99. It builds about 5x slower and adds 256 bytes of links per vector.

Scalar quantization (compressed HNSW). enable_sq(8) maps each dimension's trained range onto 256 evenly spaced values, so a vector takes dim bytes instead of 4 * dim (on SIFT, 128 instead of 512); enable_sq(4) uses 16 values packed two per byte (64 bytes on SIFT). Distances become approximate, so wrap the index in a RefineIndex to re-rank with exact ones: k_factor=2 is enough for 8 bits; 4 bits needs 4 on SIFT and 8 on GloVe word embeddings. Values outside the training sample's range are clamped, so train on a representative sample (or on the vectors themselves).

index = qann.HNSWIndex(128)
index.enable_sq(8)
index.train(x)
refine = qann.RefineIndex(index, k_factor=2)
refine.add(x)

On SIFT1M with the machine busy, at ef_search=256: 8 bits + RefineIndex(k_factor=2) matched the float graph's recall (0.996) at 1.4x its queries per second on all cores (15.8k vs 11.4k) and 1.7x on one thread, and 4 bits + k_factor=4 did nearly the same (0.995, 15.6k). The RefineIndex keeps full vectors too, so the memory saving needs the plain SQ index: 8 bits alone reaches 0.986, 4 bits alone only 0.86, so 4 bits really needs re-ranking. 4 bits is no faster than 8, since searches wait on memory latency rather than bytes moved; choose it to halve code memory. On GloVe-100 (cosine), 8 bits + k_factor=2 again matched the floats exactly, while 4 bits alone reached only 0.70 and needed k_factor=8.

RefineIndex(base, k_factor=10, *, custom_ids=False)

Wraps a trained, empty approximate index created without custom ids (and never compacted) (give the RefineIndex its own custom_ids instead). Each query fetches k * k_factor candidates from base, then re-ranks them with exact distances against full-precision copies of the vectors. Add, remove and compact through the RefineIndex: while it wraps a base, calling add, remove or compact on the base itself raises ValueError (it would put the two out of step), and a base can only be wrapped by one RefineIndex at a time. Tuning the base (nprobe, pq_distance) is fine. k_factor is a property and can be changed at any time, and the read-only base property returns the wrapped index, e.g. to change nprobe.

This is the usual way to combine PQ's speed with near-exact ranking. It keeps the original vectors in memory alongside the PQ codes.

InvertedIndex(*, custom_ids=False)

An exact inverted index over sparse vectors, for keyword-style retrieval: BM25 weights, or learned sparse vectors such as SPLADE, computed by you and passed in as scipy.sparse matrices (a column is a term id). It scores by dot product, so results come largest first with distance -(q . d), as with Metric.InnerProduct.

import scipy.sparse as sp

index = qann.InvertedIndex()
index.add(doc_matrix, metadata={"lang": langs})             # (n_docs, vocab) CSR
ids, dists = index.batch_query(query_matrix, 10, where="lang = 'en'")
ids, dists = index.query(query_matrix[[0]], 10)             # one row
  • Input: any scipy.sparse matrix or array (CSR is used directly, other formats go through tocsr()). Column indices must be in [0, 2^32); values become float32. Repeated entries in a row are summed and zeros are dropped. There is no fixed dimension.
  • Results: exact. A query reads only its own terms' posting lists and returns only documents sharing at least one nonzero term with it, so fewer than k can come back. Rows are padded with -1 / inf as for the dense indexes.
  • Shared with the dense indexes: custom ids, remove / compact, metadata and where=, allowed=, get_metadata, save / qann.load. It is a qann.SparseIndex (the sparse family), not a qann.DenseIndex, since it has no dim or metric.
  • Size: terms and postings report the distinct terms and stored (term, document) pairs. Each posting takes 8 bytes.
  • Cost: about proportional to the postings a query's terms have. With pruning on (the default), once the remaining terms can't lift an unseen document into the top k, their lists only add to documents that still could, which mostly helps small k on large corpora: 1.6x more queries per second at k = 10 on BEIR FiQA (58k documents) with BM25, about the same at k = 1000. Results are identical either way; index.pruning = False turns it off. benchmarks/sparse_bm25.py measures it.

SparseApproxIndex(*, custom_ids=False, query_terms=10, list_depth=10000, k_factor=10)

Approximate search over sparse vectors, for learned sparse vectors such as SPLADE whose queries have dozens of terms, where exact search reads a large share of the index for every query. Same input, results and shared features as InvertedIndex.

index = qann.SparseApproxIndex()            # query_terms=10, list_depth=10000, k_factor=10
index.add(doc_matrix)
ids, dists = index.batch_query(query_matrix, 10)
index.list_depth = 2000                     # faster, lower recall; applies to later queries
  • How it searches: a query uses only its query_terms largest values, reads only the list_depth heaviest entries of each of those terms' posting lists, and re-scores the k * k_factor best candidates exactly against the whole query. Every result's distance is its exact one (the same as InvertedIndex returns); only which documents are found is approximate. 0 for query_terms or list_depth means all.
  • Memory: posting lists kept heaviest first, plus a forward index (each document's own terms) for the re-scoring: about twice InvertedIndex.
  • Speed: on SPLADE vectors of 1M MS MARCO passages (big-ann-benchmarks sparse track), one thread on an M3, against exact search at 118 queries per second:
query_terms / list_depth / k_factor recall@10 queries per second
10 / 2,000 / 50 0.907 2,477 (21x)
10 / 10,000 / 10 (the defaults) 0.962 1,621 (14x)
10 / 10,000 / 50 0.989 1,183 (10x)
20 / 10,000 / 50 0.994 653 (5.5x)

benchmarks/sparse_bigann.py --approx measures it. The settings are properties, saved with the index, and can change at any time.

make_index(type, dim, opts=IndexOptions())

Builds an index from a type and an options object, returning a FlatIndex, IVFIndex, HNSWIndex, InvertedIndex or SparseApproxIndex:

opts = qann.IndexOptions()
opts.nlist, opts.nprobe, opts.pq_subspaces = 1024, 16, 16
ivf = qann.make_index(qann.IndexType.IVF, 128, opts)
ivf.train(x)

IndexType is Flat, IVF, HNSW, Inverted or SparseApprox (sparse; dim is unused, pass 0). IndexOptions fields and defaults: custom_ids (False) for all, metric (Metric.L2) for dense, pruning (True) for Inverted, query_terms (10), list_depth (10000) and k_factor (10) for SparseApprox, plus capacity (Flat initial reserve, 1024), nlist (100), nprobe (10), pq_subspaces (0, meaning no PQ), pq_centroids (256) and ivf_sq_bits (0; 8 or 4 calls enable_sq()) for IVF, and hnsw_m (16), ef_construction (200), ef_search (64) and hnsw_sq_bits (0 for floats; 8 or 4 calls enable_sq()) for HNSW.

Metrics

Every index takes metric=qann.Metric.L2 (the default), metric=qann.Metric.Cosine or metric=qann.Metric.InnerProduct, fixed at construction and saved with the index. Most text and image embedding models are meant to be compared by cosine similarity, so use Cosine for them:

index = qann.HNSWIndex(384, metric=qann.Metric.Cosine)
index.add(embeddings)
ids, dists = index.batch_query(query_embeddings, 10)   # dists = 1 - cosine similarity
  • Distances are 1 - cosine similarity, from 0 (same direction) to 2 (opposite), matching qann.cosine_distance. Similarity is 1 - dist.
  • Vector length doesn't matter: each vector is scaled to length 1 when it's added and each query when it's run, so you don't need to normalize anything yourself. A zero vector has no direction, so adding or querying one raises ValueError.
  • It works with every index, including PQ, fast-scan and HNSW. IVF learns its clusters on the scaled vectors, and a RefineIndex uses its base's metric. index.metric reports it.

Inner product

Use InnerProduct when a vector's length carries meaning, so the largest dot product q . x should come first rather than the closest direction: recommendation (user and item factors, where an item's length reflects its popularity), multi-vector retrieval such as ColBERT or MUVERA, attention keys, and embedding models trained with a dot-product loss.

index = qann.HNSWIndex(64, metric=qann.Metric.InnerProduct)
index.add(item_factors)
ids, dists = index.batch_query(user_factors, 10)   # dists = -(q . x)
  • Distances are -(q . x), so results stay smallest first; the inner product is -dist. Vectors are used as given: nothing is normalized, and zero vectors are allowed.
  • It works with every index: Flat, IVF (floats, SQ, PQ, fast-scan, rotation), HNSW (floats and SQ) and RefineIndex. PQ can only use ADC: setting pq_distance to SDC raises ValueError, and precomputed_tables reads False, since one lookup table per query already serves every list.
  • HNSW answers queries by -(q . x) but links its graph as if every stored vector had one more coordinate, sqrt(M^2 - |x|^2) with M the largest stored length. That turns inner product search into L2 search for the links, which matters a lot: at ef_search=64, recall@10 was 0.96 instead of 0.05 on ann-benchmarks' Last.fm factors and 0.50 instead of 0.05 on VIBE's llama-128-ip, against linking by raw inner product as hnswlib does. Only the length of each vector is stored for it, and M simply grows when longer vectors are added later.
  • IVF clusters by L2 as usual and probes the lists whose centroids have the largest q . c. With exact lists and nprobe=64, recall@10 was 0.98 on Last.fm, 0.88 on VIBE's yi-128-ip and 0.84 on llama-128-ip; with fast-scan PQ (rotated, dim / 2 subspaces) and RefineIndex(k_factor=8), 0.91, 0.81 and 0.73.

Custom ids and deleting

By default an index numbers vectors 0, 1, 2, ... in the order they're added. To use your own ids (database keys, document ids, ...), create the index with custom_ids=True and pass one id per vector:

index = qann.FlatIndex(128, custom_ids=True)
index.add(vectors, ids=doc_ids)     # doc_ids: one int per row
ids, dists = index.batch_query(queries, 10)   # ids are your doc_ids

index.remove([doc_ids[0], doc_ids[1]])        # returns 2
  • Ids can be a NumPy array of any integer type, a list or a range. They must be >= 0 (-1 marks missing results) and unique within the index; a bad batch raises an error and adds nothing.
  • Every add must pass ids on a custom_ids=True index, and an index created without it refuses them.
  • remove(ids) works with or without custom ids. Unknown and already deleted ids are skipped, so deleting twice is harmless. A deleted id can be added again, with a new vector.
  • Deleted vectors keep their memory until compact(): queries skip them, but they still take space and scan time. compact() frees them in one pass, keeping every id and every query result the same, so call it after a large batch of deletes. It doesn't retrain anything. (HNSWIndex is the exception: it rebuilds its graph, so ids stay the same but results can shift slightly.)
  • Default ids never repeat. New vectors continue after the highest id ever assigned, even after deletes and compact() (delete ids 0 to 9, compact, add one more, and it gets id 10). add() returns the ids it assigned, so keep those rather than counting. For data that changes (updates, deletes, re-adds), prefer custom_ids=True with ids from your own data.
  • For make_index, set IndexOptions.custom_ids.

To search only some of the vectors ("nearest products in this category", "documents this user may see"), pass the ids that may be returned as allowed:

in_stock = product_ids[stock > 0]   # computed from your own metadata
ids, dists = index.batch_query(queries, 10, allowed=in_stock)
  • Ids from your own data: keep metadata wherever you already do and turn the condition into a list of ids, or store it in the index and filter with where (see Metadata). allowed takes the same kinds of ids as remove (any integer array, list or range), with or without custom ids, and one set applies to every row of a batch_query.
  • Unknown and deleted ids in allowed are ignored. An empty set returns no results; fewer than k allowed vectors gives fewer results (rows padded with -1 / inf).
  • Filtering happens during the search, not after it, so a selective filter still returns k results instead of the few that happen to be in an unfiltered top k.
  • Works on every index. FlatIndex scores only the allowed vectors. IVFIndex skips the others while scanning, and if its nprobe lists hold fewer than k allowed vectors it keeps probing the next nearest lists until it has k. HNSWIndex walks its graph through every node but returns only allowed ones; when few are allowed (below about sqrt(ef_search * M * size), ~3% of SIFT1M), walking the graph would cost more than scoring the allowed vectors directly, so it does that instead, which is also exact. A RefineIndex passes the filter to its base and re-ranks as usual.
  • Recall by index: FlatIndex is exact. HNSWIndex narrows its search width as the filter narrows (max(k, ef_search * sqrt(s)) for a fraction s allowed): a filtered beam finds more of its true neighbors than an unfiltered one, so at the full ef_search it overshoots unfiltered recall at several times the cost. With the scaling, filtered recall stays at about the unfiltered level (SIFT1M at ef_search=64: 0.96 to 0.975 from 2% to 50% allowed, 0.953 unfiltered; GloVe-100 within 0.01 of unfiltered) and queries are 1.3 to 3.7 times faster (10% allowed: 405 to 143 us). Set filter_ef_scaling = False for the old, slower, higher-recall behavior (not saved with the index). Below about 2% allowed (2.5% at ef_search=128) HNSW scores the allowed vectors directly, which is exact. IVFIndex probes more lists as the filter narrows (about 0.75 * nprobe * s^(-1/3) for a fraction s allowed), which keeps recall near an unfiltered query's: on SIFT1M at nprobe=16, 0.92 to 0.96 from 0.2% to 20% allowed (0.93 unfiltered), where stopping at the first 10 allowed vectors gave 0.60 to 0.88. That costs speed: 1.3 to 5.7 times fewer queries per second than without it; set filter_probe_scaling = False for the faster, lower-recall behavior (not saved with the index). It helps most on hard data (GloVe-100: filtered recall 0.50-0.75 off, 0.72-0.78 on, 0.81 unfiltered); if filtered recall is already high on your data (fashion-mnist: 0.98 or more either way), turning it off is safe and up to 1.9 times faster. Under very narrow filters IVF scores the allowed vectors directly instead: a plain or SQ IVFIndex at up to 3 * nlist allowed (exact for plain; a 6-byte-per-vector lookup), and a RefineIndex over IVF at up to 6 * nlist (exact, from its raw copies). A PQ IVFIndex doesn't, so wrap it in a RefineIndex for narrow filters. Full numbers are in ARCHITECTURE.md.
  • Cost is a pass over allowed per call (to look up each id), plus one bit per vector in the index.

Metadata

Store attributes with each vector and filter on them directly:

index.add(vectors, metadata={"category": categories, "price": prices, "year": years})

f = qann.field
ids, dists = index.batch_query(queries, 10, where=(f("category") == "shoes") & (f("price") < 50))
ids, dists = index.batch_query(queries, 10, where=f("year").isin([2024, 2025]) | f("price").is_null())

# or as a SQL WHERE clause, with values passed separately
ids, dists = index.batch_query(queries, 10, where="category = 'shoes' AND price < 50")
ids, dists = index.batch_query(queries, 10, where=("user_id = ? AND year IN (?, ?)", [uid, 2024, 2025]))
  • Columns are named and hold ints, floats or strings, one value per vector, as NumPy arrays or lists (bool arrays become ints). A column's type is fixed by the first add that has it, or by add_metadata_column. Ints are accepted into a float column (so whole-number prices work; ints beyond 2^53 are rounded to the nearest float), but floats into an int column raise ValueError.
  • Missing values: None in a column's values is a null. A column added later is null for the vectors already in the index, and an add that leaves a column out stores null for its vectors. Nulls fail every comparison and isin, so ~(f("price") < 50) includes vectors with no price; use f("price").is_null() to select or exclude them.
  • Filters: ==, !=, <, <=, >, >=, .isin([...]), .between(lo, hi) (both ends included) and .is_null(), combined with & (and), | (or) and ~ (not). Parenthesize comparisons, since & binds tighter than <: (f("a") < 1) & (f("b") == 2). Python's and, or, not and chained comparisons such as 0 < f("a") < 5 raise TypeError instead of silently dropping part of the filter. For a long chain built in code, use qann.any_of(filters) / qann.all_of(filters): adding terms one | at a time copies the chain each step, so it gets slow past a few thousand terms. For one column against many values, f("price").isin(values) is better still: it's one condition, so it's scanned once per query instead of once per value.
  • SQL strings: where also takes a SQL WHERE clause: =, != / <>, <, <=, >, >=, [NOT] IN (...), IS [NOT] NULL, [NOT] BETWEEN x AND y, AND, OR, NOT and parentheses, with SQL's precedence and case-insensitive keywords. Strings are '...' ('' for a quote); quote column names that are keywords or contain non-ASCII characters as "...". Numbers that overflow, such as 1e400, raise ValueError. LIKE, functions, arithmetic and comparing two columns aren't supported. A syntax error raises ValueError with its column. Filters can nest up to 256 levels of parentheses or NOT; long AND / OR chains don't count toward that. A string may have up to 1024 conditions (comparisons, IN, IS NULL, BETWEEN; an IN list counts once, however many values it has), so filter strings from users can't make a query scan its columns thousands of times; qann.set_max_filter_conditions(n) changes the limit for the whole process. Filters built with qann.field aren't limited.
  • Parameters: pass values as (sql, [values]) for ? or (sql, {name: value}) for :name instead of pasting them into the string. They are bound after parsing, so user input can't change the filter ("1 OR 1 = 1" is just a value). qann.where(sql, params) returns the filter, to reuse it or combine it with qann.field filters through &, | and ~.
  • NOT vs ~: SQL's NOT excludes nulls, as in SQL: NOT price < 50 doesn't match vectors with no price, while ~(f("price") < 50) does.
  • Types: ints and floats compare as numbers; strings compare with strings (< is alphabetical). Filtering on a column the index doesn't have, or comparing a number column with a string, raises ValueError.
  • Same search as allowed: a where filter becomes the set of matching vectors and runs the filtered search for every index, with the same recall behavior. With both where and allowed, only vectors passing both are returned.
  • Cost: each query call scans the columns its filter names once, shared by every row of a batch_query: about 0.4 ms per condition per million vectors. Strings are stored once per distinct value, plus 4 bytes per vector; ints and floats take 8 bytes per vector; each column also takes 1 byte per vector for its null flag.
  • Reading back: index.get_metadata(ids) returns {name: values} with one value per id, for every column in the order they were added; columns=["price", "color"] picks and orders them. Int and float columns come back as NumPy masked arrays (int64 / float64, masked where the value is null), so ints stay exact and .filled(x) or .compressed() handle nulls; string columns come back as lists with None for null. Unknown, deleted or repeated ids, and columns the index doesn't have, raise ValueError. With update_metadata this gives read-modify-write: stock = index.get_metadata(ids, columns=["stock"])["stock"], then index.update_metadata(ids, {"stock": (stock - 1).tolist()}).
  • Updating: index.update_metadata(ids, {"price": [10, None]}) overwrites the given columns for those vectors in place (no re-insert), with one value per id; other columns keep their values. A column the index doesn't have raises ValueError, so a typo can't create one.
  • New columns for existing vectors: index.add_metadata_column("in_stock", "int", default=1) gives every vector that value; index.add_metadata_column("tag", "str", ids=tagged, values=tags) sets those vectors (one value per id, None for null) and leaves the rest null, in one step; with neither, every vector is null. default can't be combined with ids / values (TypeError). Later adds that leave the column out store null. drop_metadata_column(name) removes a column, freeing its name; compact() never removes columns, even ones left with no values. Unknown, deleted or repeated ids raise ValueError, and a failed call changes nothing. Strings no vector uses any more are freed by compact(). Re-adding a removed id doesn't bring back its old metadata: pass it again with the add.
  • Kept with the index: remove and compact keep metadata with its vector, and save / qann.load store it. A RefineIndex holds the metadata itself; add through it as usual.

Saving and loading

Any index can be saved to a file and loaded back, including its training, so it doesn't have to be retrained or re-added each run:

index.save("vectors.qann")          # str or pathlib.Path

index = qann.load("vectors.qann")   # returns the saved class, e.g. IVFIndex or InvertedIndex
ids, dists = index.batch_query(queries, 10)
  • Everything is restored: vectors, ids (including custom ids), deletions, IVF centroids, PQ codebooks, HNSW graphs and SQ ranges, the metric and settings such as nprobe, pq_distance, precomputed_tables, ef_search and k_factor. A loaded index gives identical results and keeps accepting add and remove.
  • A RefineIndex file contains its base index, so loading one restores both. Tune the base through loaded.base, e.g. loaded.base.nprobe = 32.
  • Saving is safe to interrupt: the file is written next to the target as <path>.tmp and renamed into place when complete, so a failed save never leaves a partial file or destroys an existing one.
  • Bad files raise RuntimeError: a missing, truncated or corrupted file, or one that isn't a QaNN index, is rejected instead of loading garbage.
  • Files are versioned. Newer releases keep loading files saved by older ones (files from 0.1.0a4, 0.1.0a6, 0.1.0a7, 0.1.0a10, 0.1.0a11 and 0.1.0a12 still load). A file saved by a newer release than the one loading it raises RuntimeError rather than being misread.

PQCodebook on its own can't be saved yet; save the IVFIndex that uses it instead.

Product quantization

PQCodebook(dim, num_subspaces, centroids_per_subspace=256) exposes the quantizer used inside IVFIndex for direct use.

Method Description
train(data, max_iters=25, seed=None) Run k-means in each subspace over an (n, dim) array.
encode(vec) Encode a (dim,) vector into a uint8 array of num_subspaces codes.
compute_adc_table(query) Precompute distances from a (dim,) query to every centroid; reuse it across many codes.
distance_adc(table, code) Approximate squared L2 between the table's query and a code (asymmetric).
distance_sdc(query_code, code) Approximate squared L2 between two codes (symmetric).
dim(), num_subspaces(), centroids_per_subspace() Shape of the codebook.
pq = qann.PQCodebook(128, 16)
pq.train(x, seed=1)
codes = [pq.encode(v) for v in x[:1000]]
table = pq.compute_adc_table(x[0])
dists = [pq.distance_adc(table, c) for c in codes]

Utilities

Function Description
l2_distance(a, b) Row-wise squared L2 distance between two (m, dim) arrays; returns shape (m,).
cosine_distance(a, b) Row-wise cosine distance (1 minus cosine similarity) between two (m, dim) arrays.
set_num_threads(n) Threads used for batch queries, adding with PQ and PQ training. 0 (default) uses every core; 1 runs serially.
num_threads() Thread count currently in effect.
__version__ Installed version.

Threads and type hints

  • Thread-safe. Queries, adds and training release Python's GIL while they run, so other Python threads keep running, and several threads can query the same index at once. An add, train or setting change waits for running queries on that index to finish (and blocks new ones until it's done), so concurrent use never sees a half-updated index. This also covers the base of a RefineIndex, e.g. ivf.nprobe = 32 while queries run through the RefineIndex wrapping ivf.
  • Typed. Wheels include type stubs, so editors autocomplete the API and type checkers such as mypy and Pylance/pyright check calls to it.

More

See ARCHITECTURE.md for the design, a file-by-file reference and SIFT1M benchmarks.

Metadata

Release files for qann 0.1.0a24

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for qann 0.1.0a24
File Size Uploaded Provenance Actions
qann-0.1.0a24.tar.gz 2.9 MB With provenance Details for qann-0.1.0a24.tar.gz

Built distributions (wheels)

Table of built distributions (wheels) for qann 0.1.0a24
File Actions
qann-0.1.0a24-cp314-cp314-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl CPython 3.14 CPython 3.14 Linux glibc 2.28+ x86-64, Linux glibc 2.27+ x86-64 Details for qann-0.1.0a24-cp314-cp314-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl
qann-0.1.0a24-cp314-cp314-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl CPython 3.14 CPython 3.14 Linux glibc 2.28+ ARM64, Linux glibc 2.27+ ARM64 Details for qann-0.1.0a24-cp314-cp314-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl
qann-0.1.0a24-cp314-cp314-macosx_11_0_x86_64.whl CPython 3.14 CPython 3.14 macOS 11.0+ x86-64 Details for qann-0.1.0a24-cp314-cp314-macosx_11_0_x86_64.whl
qann-0.1.0a24-cp314-cp314-macosx_11_0_arm64.whl CPython 3.14 CPython 3.14 macOS 11.0+ ARM64 Details for qann-0.1.0a24-cp314-cp314-macosx_11_0_arm64.whl
qann-0.1.0a24-cp313-cp313-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl CPython 3.13 CPython 3.13 Linux glibc 2.28+ x86-64, Linux glibc 2.27+ x86-64 Details for qann-0.1.0a24-cp313-cp313-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl
qann-0.1.0a24-cp313-cp313-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl CPython 3.13 CPython 3.13 Linux glibc 2.28+ ARM64, Linux glibc 2.27+ ARM64 Details for qann-0.1.0a24-cp313-cp313-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl
qann-0.1.0a24-cp313-cp313-macosx_11_0_x86_64.whl CPython 3.13 CPython 3.13 macOS 11.0+ x86-64 Details for qann-0.1.0a24-cp313-cp313-macosx_11_0_x86_64.whl
qann-0.1.0a24-cp313-cp313-macosx_11_0_arm64.whl CPython 3.13 CPython 3.13 macOS 11.0+ ARM64 Details for qann-0.1.0a24-cp313-cp313-macosx_11_0_arm64.whl
qann-0.1.0a24-cp312-cp312-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl CPython 3.12 CPython 3.12 Linux glibc 2.28+ x86-64, Linux glibc 2.27+ x86-64 Details for qann-0.1.0a24-cp312-cp312-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl
qann-0.1.0a24-cp312-cp312-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl CPython 3.12 CPython 3.12 Linux glibc 2.28+ ARM64, Linux glibc 2.27+ ARM64 Details for qann-0.1.0a24-cp312-cp312-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl
qann-0.1.0a24-cp312-cp312-macosx_11_0_x86_64.whl CPython 3.12 CPython 3.12 macOS 11.0+ x86-64 Details for qann-0.1.0a24-cp312-cp312-macosx_11_0_x86_64.whl
qann-0.1.0a24-cp312-cp312-macosx_11_0_arm64.whl CPython 3.12 CPython 3.12 macOS 11.0+ ARM64 Details for qann-0.1.0a24-cp312-cp312-macosx_11_0_arm64.whl
qann-0.1.0a24-cp311-cp311-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl CPython 3.11 CPython 3.11 Linux glibc 2.28+ x86-64, Linux glibc 2.27+ x86-64 Details for qann-0.1.0a24-cp311-cp311-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl
qann-0.1.0a24-cp311-cp311-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl CPython 3.11 CPython 3.11 Linux glibc 2.28+ ARM64, Linux glibc 2.27+ ARM64 Details for qann-0.1.0a24-cp311-cp311-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl
qann-0.1.0a24-cp311-cp311-macosx_11_0_x86_64.whl CPython 3.11 CPython 3.11 macOS 11.0+ x86-64 Details for qann-0.1.0a24-cp311-cp311-macosx_11_0_x86_64.whl
qann-0.1.0a24-cp311-cp311-macosx_11_0_arm64.whl CPython 3.11 CPython 3.11 macOS 11.0+ ARM64 Details for qann-0.1.0a24-cp311-cp311-macosx_11_0_arm64.whl
qann-0.1.0a24-cp310-cp310-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl CPython 3.10 CPython 3.10 Linux glibc 2.28+ x86-64, Linux glibc 2.27+ x86-64 Details for qann-0.1.0a24-cp310-cp310-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl
qann-0.1.0a24-cp310-cp310-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl CPython 3.10 CPython 3.10 Linux glibc 2.27+ ARM64, Linux glibc 2.28+ ARM64 Details for qann-0.1.0a24-cp310-cp310-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl
qann-0.1.0a24-cp310-cp310-macosx_11_0_x86_64.whl CPython 3.10 CPython 3.10 macOS 11.0+ x86-64 Details for qann-0.1.0a24-cp310-cp310-macosx_11_0_x86_64.whl
qann-0.1.0a24-cp310-cp310-macosx_11_0_arm64.whl CPython 3.10 CPython 3.10 macOS 11.0+ ARM64 Details for qann-0.1.0a24-cp310-cp310-macosx_11_0_arm64.whl

Total release size: 97.3 MB

Release files / qann-0.1.0a24.tar.gz

Download URL qann-0.1.0a24.tar.gz
Size 2.9 MB
Tags Source
SHA-256 checksum
How to use checksums
987db2a3af368cbce486359fcdc6c7ac414190ec335fa9f7634b6dcaea5064ab
BLAKE2b-256 checksum
How to use checksums
e80248b7a03d6661ecc12dc597043af11c44f5fd477ca5223666f0792e1c1dd5
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 1, 2026.

Transparency log

Release files / qann-0.1.0a24-cp314-cp314-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl

Download URL qann-0.1.0a24-cp314-cp314-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl
Size 12.6 MB
Tags CPython 3.14 Linux glibc 2.27+ x86-64 Linux glibc 2.28+ x86-64
SHA-256 checksum
How to use checksums
6e44fc8ce0545566b82c78a1b6afd1adfa4aa3808e32bf4a82deaf3d1f14e554
BLAKE2b-256 checksum
How to use checksums
88479a15d7dabdff768d9841d4a10fca4b1f5756905af0c0f86ef658eacc2fde
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 1, 2026.

Transparency log

Release files / qann-0.1.0a24-cp314-cp314-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl

Download URL qann-0.1.0a24-cp314-cp314-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl
Size 5.5 MB
Tags CPython 3.14 Linux glibc 2.27+ ARM64 Linux glibc 2.28+ ARM64
SHA-256 checksum
How to use checksums
ff2939b7abfd2a25259b2f2f919915ad6e2ee56dd721913dafdfb618bf91e648
BLAKE2b-256 checksum
How to use checksums
efa2465036a9f526dd7a12025485e2be89c7307cb85ed8a2a9aa45cb6226ed06
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 1, 2026.

Transparency log

Release files / qann-0.1.0a24-cp314-cp314-macosx_11_0_x86_64.whl

Download URL qann-0.1.0a24-cp314-cp314-macosx_11_0_x86_64.whl
Size 431.7 kB
Tags CPython 3.14 macOS 11.0+ x86-64
SHA-256 checksum
How to use checksums
3124dec59c700116b8a0b9c9c8f8b0632fd578f9eb3b11a6176a6dd4d22a7a95
BLAKE2b-256 checksum
How to use checksums
53e378e003ab732f10fc682f33a320007b325fadfe20bb17f976e6196cb804b2
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 1, 2026.

Transparency log

Release files / qann-0.1.0a24-cp314-cp314-macosx_11_0_arm64.whl

Download URL qann-0.1.0a24-cp314-cp314-macosx_11_0_arm64.whl
Size 370.8 kB
Tags CPython 3.14 macOS 11.0+ ARM64
SHA-256 checksum
How to use checksums
d6423e1003694177d23395d64a6ca564bbf61838fdee19f7828d3022a19b67ba
BLAKE2b-256 checksum
How to use checksums
77da4b287fe2a4c5f1356416bfb4f960f3578749043d7590cdf2f3aed3719e92
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 1, 2026.

Transparency log

Release files / qann-0.1.0a24-cp313-cp313-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl

Download URL qann-0.1.0a24-cp313-cp313-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl
Size 12.6 MB
Tags CPython 3.13 Linux glibc 2.27+ x86-64 Linux glibc 2.28+ x86-64
SHA-256 checksum
How to use checksums
661aa3bcc42cac154f0224c60d501df15d6a643e03eacca9ba39da5f820f22de
BLAKE2b-256 checksum
How to use checksums
4e39479f9cf47317cb31f8ad3764af4f1790c9ee395ad0603b0699459a3a2369
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 1, 2026.

Transparency log

Release files / qann-0.1.0a24-cp313-cp313-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl

Download URL qann-0.1.0a24-cp313-cp313-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl
Size 5.5 MB
Tags CPython 3.13 Linux glibc 2.27+ ARM64 Linux glibc 2.28+ ARM64
SHA-256 checksum
How to use checksums
299a4c884c263f0edb980c2943cc0900a4368dc3240e0424e5aaf4046677b626
BLAKE2b-256 checksum
How to use checksums
251f6db5e116520d3ce196fbb5c993387a46f586a21569189a685df2309ec710
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 1, 2026.

Transparency log

Release files / qann-0.1.0a24-cp313-cp313-macosx_11_0_x86_64.whl

Download URL qann-0.1.0a24-cp313-cp313-macosx_11_0_x86_64.whl
Size 431.6 kB
Tags CPython 3.13 macOS 11.0+ x86-64
SHA-256 checksum
How to use checksums
98e114c783d08bc6575ccab493ef858ea0c782ec2ed5176fac588614284c61ad
BLAKE2b-256 checksum
How to use checksums
0a061e9669ee2d46ddcdb5bd0026b314acf00cfb5c02a1c8f800126070ce5ce1
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 1, 2026.

Transparency log

Release files / qann-0.1.0a24-cp313-cp313-macosx_11_0_arm64.whl

Download URL qann-0.1.0a24-cp313-cp313-macosx_11_0_arm64.whl
Size 370.7 kB
Tags CPython 3.13 macOS 11.0+ ARM64
SHA-256 checksum
How to use checksums
0a2879b098ef45ce54718f8d900f91d64b1c7da68c101d8adb6a1e9ef6a42efa
BLAKE2b-256 checksum
How to use checksums
dc02be4c9a6346d2227c4ff1c4e215a66ebe471750bc2090073e7503a556f49b
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 1, 2026.

Transparency log

Release files / qann-0.1.0a24-cp312-cp312-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl

Download URL qann-0.1.0a24-cp312-cp312-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl
Size 12.6 MB
Tags CPython 3.12 Linux glibc 2.27+ x86-64 Linux glibc 2.28+ x86-64
SHA-256 checksum
How to use checksums
5acc3d4a03f48f642bca2ddd4a26a7a9af8df9e81bef4ce7c2b5afb476ca90bf
BLAKE2b-256 checksum
How to use checksums
05d26af7777234924f33ad6fb05d500b1305dbd12adaf44e4d36bf99552f3510
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 1, 2026.

Transparency log

Release files / qann-0.1.0a24-cp312-cp312-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl

Download URL qann-0.1.0a24-cp312-cp312-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl
Size 5.5 MB
Tags CPython 3.12 Linux glibc 2.27+ ARM64 Linux glibc 2.28+ ARM64
SHA-256 checksum
How to use checksums
edddf804593c7390a9288677281292e7103e6bed1bef757c4d4e1f018158588e
BLAKE2b-256 checksum
How to use checksums
74b8ca5e5ed5c93dc846a7b50f1b8e5ebf037c693b3884d1d495ceff31187286
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 1, 2026.

Transparency log

Release files / qann-0.1.0a24-cp312-cp312-macosx_11_0_x86_64.whl

Download URL qann-0.1.0a24-cp312-cp312-macosx_11_0_x86_64.whl
Size 431.7 kB
Tags CPython 3.12 macOS 11.0+ x86-64
SHA-256 checksum
How to use checksums
eba55b66d39c8f8b47fc482c510089498b28c61072665af7801ff5bec9fabaad
BLAKE2b-256 checksum
How to use checksums
d83b3808aa55e6bbc165adb676df76fb5bc3dc97b5f80a883e3dbd5b1aee2574
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 1, 2026.

Transparency log

Release files / qann-0.1.0a24-cp312-cp312-macosx_11_0_arm64.whl

Download URL qann-0.1.0a24-cp312-cp312-macosx_11_0_arm64.whl
Size 370.6 kB
Tags CPython 3.12 macOS 11.0+ ARM64
SHA-256 checksum
How to use checksums
076acca67cd2d19ea15baf35710707abb4ba79fd53f3d69ac1be94b013d33da8
BLAKE2b-256 checksum
How to use checksums
9401b4225364f26a8323467fe2b91e625de2cca23ac54f97245565eb87f5b88e
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 1, 2026.

Transparency log

Release files / qann-0.1.0a24-cp311-cp311-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl

Download URL qann-0.1.0a24-cp311-cp311-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl
Size 12.6 MB
Tags CPython 3.11 Linux glibc 2.27+ x86-64 Linux glibc 2.28+ x86-64
SHA-256 checksum
How to use checksums
d093dd97f1ea3f7ee99e5045f04d20c53709953e94ebacf6d1d32eb23a5b818f
BLAKE2b-256 checksum
How to use checksums
f7ca178e37163ca35a2b81d0ad9788a865a800df3c694d035d463522c32035c9
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 1, 2026.

Transparency log

Release files / qann-0.1.0a24-cp311-cp311-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl

Download URL qann-0.1.0a24-cp311-cp311-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl
Size 5.5 MB
Tags CPython 3.11 Linux glibc 2.27+ ARM64 Linux glibc 2.28+ ARM64
SHA-256 checksum
How to use checksums
1b351e7c420e43a22c59d01b567614a7f30adcdeb0ac6d1379c8e5cc2fd5e38e
BLAKE2b-256 checksum
How to use checksums
9e99128d1ce1af383d4f1851905aaf41d5f048201cd919c65d2ee2fb9d43cb12
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 1, 2026.

Transparency log

Release files / qann-0.1.0a24-cp311-cp311-macosx_11_0_x86_64.whl

Download URL qann-0.1.0a24-cp311-cp311-macosx_11_0_x86_64.whl
Size 431.7 kB
Tags CPython 3.11 macOS 11.0+ x86-64
SHA-256 checksum
How to use checksums
d3d41ce9e0745c0c4ff0ff0d9672b4b706822a12fafb866bdb796983600ec399
BLAKE2b-256 checksum
How to use checksums
a8e268438893c1e0f54c79bcc1d5611750e5e2d2e2061f293bbb1a473e11e680
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 1, 2026.

Transparency log

Release files / qann-0.1.0a24-cp311-cp311-macosx_11_0_arm64.whl

Download URL qann-0.1.0a24-cp311-cp311-macosx_11_0_arm64.whl
Size 371.3 kB
Tags CPython 3.11 macOS 11.0+ ARM64
SHA-256 checksum
How to use checksums
5820ae6c60d4dbc0ff36dbdad2430bf3e51a9fdcd959b56a2f06df96ef29e8aa
BLAKE2b-256 checksum
How to use checksums
8248595d20423cab9c9b791020e98a5d37ab09ef1294c0468235455d79b4b592
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 1, 2026.

Transparency log

Release files / qann-0.1.0a24-cp310-cp310-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl

Download URL qann-0.1.0a24-cp310-cp310-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl
Size 12.6 MB
Tags CPython 3.10 Linux glibc 2.27+ x86-64 Linux glibc 2.28+ x86-64
SHA-256 checksum
How to use checksums
1a98d557080d7a8b8efd2eed4822e7571ff0cf0ab6055bb55b45660bb9eabf9a
BLAKE2b-256 checksum
How to use checksums
d4d4355f352dc92e2c82040b55704977b5ec323e58dbbfa4b0058d4b9f8db784
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 1, 2026.

Transparency log

Release files / qann-0.1.0a24-cp310-cp310-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl

Download URL qann-0.1.0a24-cp310-cp310-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl
Size 5.5 MB
Tags CPython 3.10 Linux glibc 2.27+ ARM64 Linux glibc 2.28+ ARM64
SHA-256 checksum
How to use checksums
1ec140a1350d3eccc7cace9b2188b498dd81227ca8bd038c6b91f4a0c9454d58
BLAKE2b-256 checksum
How to use checksums
91bf3ddf9ff57d15809b04abe3c80934349a5fcf4832911418400e7ccae9cec5
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 1, 2026.

Transparency log

Release files / qann-0.1.0a24-cp310-cp310-macosx_11_0_x86_64.whl

Download URL qann-0.1.0a24-cp310-cp310-macosx_11_0_x86_64.whl
Size 431.5 kB
Tags CPython 3.10 macOS 11.0+ x86-64
SHA-256 checksum
How to use checksums
545e85b9b64a1f5993d8a7acc13ee8a93ade55e8bc7b3e8ec7ad268e4853ad7a
BLAKE2b-256 checksum
How to use checksums
7eedd69f121ab8041929cb9f648c749504ff948633e137f99492fce2c5736dc8
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 1, 2026.

Transparency log

Release files / qann-0.1.0a24-cp310-cp310-macosx_11_0_arm64.whl

Download URL qann-0.1.0a24-cp310-cp310-macosx_11_0_arm64.whl
Size 371.1 kB
Tags CPython 3.10 macOS 11.0+ ARM64
SHA-256 checksum
How to use checksums
852667cb3fe9bf8935d786f125f2f2f8d839ee0595e24e3eaf9460ed6c3d5e7b
BLAKE2b-256 checksum
How to use checksums
63d0869c2726d892a700c139a1b6f7809e806345401a65cfe51821e0f267fbaa
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 1, 2026.

Transparency log

Security for qann 0.1.0a24

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

All 21 files in this release have been uploaded with provenance metadata.

Select from 21 files

Attestation for qann-0.1.0a24.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 1, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Release history Release notifications | RSS feed

This release
Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page