dejanseo/200k-wiki-data-embeddings-gemma

200k Wikidata Embeddings (EmbeddingGemma)

Text embeddings for ~200,000 Wikidata entities, generated from each entity's English label with Google's open google/embeddinggemma-300m model. Vectors are 768-dimensional and unit length (the model normalizes its output), so cosine similarity equals a plain dot product.

What's inside

Column	Type	Description
entity_id	string	Wikidata Q-identifier, e.g. Q42
label	string	English label, e.g. Douglas Adams
embedding	list<float32>[768]	Unit-length embedding of the label

• Rows: 199,998
• Dimensions: 768
• Normalization: L2 (every vector has norm 1.0, applied by the model)
• Source model: google/embeddinggemma-300m
• File: gemma.parquet

The entities are a sample of the full Wikidata label set (~118.7M labelled entities). Only the label string was embedded — no descriptions, aliases, or statements were used.

Usage

from datasets import load_dataset

ds = load_dataset("dejanseo/200k-wiki-data-embeddings-gemma", split="train")
print(ds[0]["entity_id"], ds[0]["label"])
print(len(ds[0]["embedding"]))  # 768

Semantic search with cosine similarity (a dot product, since vectors are unit-length):

import numpy as np

emb = np.array(ds["embedding"], dtype=np.float32)   # (199998, 768)
ids = ds["entity_id"]
labels = ds["label"]

def nearest(query_vec, k=10):
    scores = emb @ np.asarray(query_vec, dtype=np.float32)  # cosine similarity
    top = np.argsort(-scores)[:k]
    return [(ids[i], labels[i], float(scores[i])) for i in top]

To embed a new query with the same model:

from sentence_transformers import SentenceTransformer

model = SentenceTransformer("google/embeddinggemma-300m")
q = model.encode_query("science fiction author")   # matches the query prompt
hits = nearest(q, k=10)

How it was built

1. English labels were pulled from a local Wikidata label database.
2. Each label was embedded with google/embeddinggemma-300m in document mode (encode_document), which applies the model's document prompt.
3. The model returns unit-length vectors, which were written to Parquet as-is.

Companion dataset

A parallel set of embeddings for the same 199,998 entities was produced with Google's gemini-embedding-001 API model (also 768-dim, unit-length). Because the entity_id keys are identical, the two sets can be joined row-for-row for cross-model comparison.

Note on cross-model use: the two models place entities in differently-oriented spaces. Do not compare a Gemma vector directly against a Gemini vector — raw cross-model similarity is meaningless. Within a single model, similarity is well-behaved.

Licensing

Wikidata labels are released under CC0. The embedding vectors are derived outputs of google/embeddinggemma-300m and are subject to the Gemma Terms of Use. Choose and set a license: value in the metadata above that reflects how you intend to distribute the derived vectors before publishing.

Citation

If you use this dataset, please cite the source model and Wikidata:

• Google, EmbeddingGemma (google/embeddinggemma-300m)
• Wikidata (Wikimedia Foundation)