constella-zero

constella-zero is the smallest query encoder in an asymmetric retrieval family. Documents are indexed once, in the cloud, with the frozen stella-en-400M-v5-doc-onnx tower; Zero or the stronger constella-nano can then query that same 1024-dimensional index without re-encoding it. Zero is a 30,522 × 1024 int8 lookup table—not a transformer—so encoding is a gather and weighted sum.

Research preview. The registered reserved-four evaluation and broad descriptive BEIR-18 validation are pending and unspent; no result is claimed for either. The three models are registered on the preview branch below, not in an upstream FastEmbed release yet.

Usage

pip install "fastembed @ git+https://github.com/Dylancouzon/fastembed.git@constella-research-preview"
pip install qdrant-client

import numpy as np
from fastembed import TextEmbedding

NAME = "DylanCouzon/constella-zero"
query_model = TextEmbedding(NAME)
q = np.asarray(next(iter(query_model.embed(["how do mrna vaccines work?"]))))
assert q.shape == (1024,) and q.dtype == np.float32

FastEmbed fetches model.onnx and the tokenizer. Pooling and L2 normalization happen inside the graph.

The document side

DOC_NAME = "DylanCouzon/stella-en-400M-v5-doc-onnx"
doc_model = TextEmbedding(DOC_NAME)
docs = [
    "mRNA vaccines deliver messenger RNA encoding a viral antigen.",
    "The Treaty of Westphalia ended the Thirty Years' War in 1648.",
]
D = np.stack(list(doc_model.embed(docs)))
assert D.shape == (2, 1024) and D.dtype == np.float32

The document tower runs once per document; Zero runs on every query. Do not use the document model's unprompted path as a Stella query encoder.

With Qdrant

from qdrant_client import QdrantClient, models

client = QdrantClient(":memory:")
client.create_collection(
    "docs", vectors_config=models.VectorParams(size=1024, distance=models.Distance.COSINE)
)
client.upsert(
    "docs",
    points=[
        models.PointStruct(id=i, vector=D[i].tolist(), payload={"text": text})
        for i, text in enumerate(docs)
    ],
)
hits = client.query_points("docs", query=q.tolist(), limit=2).points
print(hits[0].payload["text"])

Without FastEmbed

zero_encoder.py is the NumPy/tokenizers reference path:

from huggingface_hub import snapshot_download
import sys

d = snapshot_download("DylanCouzon/constella-zero")
sys.path.insert(0, d)
from zero_encoder import ZeroQueryEncoder

enc = ZeroQueryEncoder(d, variant="int8")
q_np = enc.encode(["how do mrna vaccines work?"])
assert np.abs(q_np[0] - q).max() < 1e-5

Zero and Nano share a document space, not a retrieval-quality guarantee: they have different measured behavior, and interchangeability does not mean parity, equivalence, or a tie.

How it works

Tokenize with WordPiece, special tokens on, no prefix, and truncation at 512 tokens. A token that appears c times carries total weight sqrt(c), so repetition saturates. The graph sums the rows, divides by the weight sum, and L2-normalizes; an empty or near-zero bag falls back to normalized [CLS]. Learned token weights are folded into the rows. This is still a bag of tokens: word order, negation, and syntax are not represented.

The reported model is int8, which was within 0.00013 nDCG@10 of fp16.

Files

file	purpose	size
model.onnx	FastEmbed/ONNX Runtime; pooled normalized (batch, 1024) output	31 MB
model_tokens.onnx	token-level (batch, sequence, 1024) output	31 MB
model.npz	NumPy reference path	94 MB

Both graphs use opset 17 and standard operators. The int8 table is dequantized in-graph with one fp32 scale per row. Tokenizer metadata enforces the frozen 512-token rule and dynamic padding.

Results

The headline partition is clean-4: NFCorpus, SCIDOCS, SciFact, and TREC-COVID. ArguAna and FiQA remain beside it with † because Stella discloses training/evaluation contact with them. All values are exact-search nDCG@10.

system	NFCorpus (clean-4)	SCIDOCS (clean-4)	SciFact (clean-4)	TREC-COVID (clean-4)	ArguAna†	FiQA†
constella-zero (int8)	0.3124	0.1677	0.6101	0.5490	0.5916	0.3728
constella-nano	0.363080	0.217710	0.721097	0.787116	0.623296	0.477765
BM25	0.3180	0.1565	0.6791	0.6099	0.4878	0.2532
Stella teacher, symmetric	0.4134	0.2395	0.7796	0.8234	0.6369	0.5536

† Stella-disclosed training/evaluation contact; excluded from clean-4.

Zero did not confirmatorily beat BM25. M7 C2 was +0.0165 across all six with raw 95% interval [+0.0017, +0.0311], but its sign-flip p=0.0149 failed the Holm threshold of 0.0083. On clean-4, the descriptive contrast was -0.0311 [-0.0517, -0.0109], with Zero at 0.4098 versus BM25 at 0.4409. Superiority is UNESTABLISHED.

The deployable hybrid recommendation and registered operator of record are distinct:

Zero + BM25 fusion	prefetch	all-six macro	clean-4 macro
Qdrant DBSF	100	0.4887	0.4912
M7 convex0 (w=0.8)	1000	0.4911	0.4866

Both rows use bm25s with Lucene defaults as the lexical side, not Qdrant's own BM25, which has a fixed avg_len and its own tokenizer — DBSF normalises over returned scores, so a different lexical implementation shifts its inputs. For the DBSF row, each query's own document is excluded before the prefetch is truncated to 100 (must_not on the point id); without that filter a plain limit: 100 spends a slot on the self-match. That affects only ArguAna (1,298 of 1,406 queries) and FiQA (55), so the clean-4 figures are unchanged either way.

Use Qdrant DBSF at prefetch 100 in deployments. M7's convex0 is the registered operator of record, but Qdrant does not implement it. No confidence interval compared these observations, so neither superiority nor equivalence is established. M7 C3 likewise did not establish fusion superiority over OpenSearch: +0.0043, raw 95% interval [-0.0063, +0.0151], p=0.219.

Full per-dataset fusion rows, registered contrasts, and source traces are in the M21 benchmark ledger.

Limits

• Reserved-four and BEIR-18 evidence is pending and unspent; the six datasets do not establish broad domain coverage.
• Stella contact with ArguAna and FiQA makes all-six results secondary to clean-4.
• Zero is an English-only bag of tokens and truncates beyond 512 wordpieces.
• The document side is not cheap: the 400M-parameter Stella tower still runs at indexing time.

Costs

These are synthetic query latencies, not workload estimates. The common three-model protocol used three fresh processes per model, batch one, four CPU threads; hydration includes imports, verification, and load but excludes interpreter startup; first inference is separate; warm timing uses five warmups and twenty 20-word samples; the OS disk cache was not flushed.

model	hydration	first query	warm 20-word p50	peak RSS	model assets
constella-zero	0.2618 s	0.3529 ms	0.1119 ms	275.4 MiB	90.1 MiB
bge-small	0.6726 s	8.2401 ms	6.8400 ms	291.0 MiB	127.6 MiB
constella-nano	0.6907 s	7.6685 ms	7.2511 ms	280.9 MiB	132.3 MiB

This measures query encoders only, not retrieval, ANN, or end-to-end system latency. The asset column follows the common protocol; model.onnx itself is the 31 MB query graph listed above.

Training

The table was trained by L2 regression against Stella query embeddings over 340,850 pairs plus 220,632 query-only rows from Amazon ESCI, FEVER, HotpotQA, SQuAD, NQ Open, TriviaQA, and Mr. TyDi (English); MS MARCO was excluded. Wikipedia-derived sources retain CC BY-SA attribution; ESCI and TriviaQA are Apache-2.0.

Provenance

run_id      p35w-2m-s2500
table       a7007b1a6af120b976f093fd69ddcb5001996ec0b84b5864b4fd25d7af878abf
teacher     NovaSearch/stella_en_400M_v5 @ ffeb2b7ee715c226d4ffe5e4619f7dbb48624c20
comparator  BAAI/bge-small-en-v1.5 @ 5c38ec7c405ec4b44b94cc5a9bb96e735b38267a
preproc     prefix="" · special tokens · max_length=512 · pool_mode=sqrt
fingerprint adb24fb2e8cad66f

The weights are MIT licensed. The pinned Stella teacher/document tower and bge-small comparator are also MIT and are attributed above.