Embeddings Specialist

Turn text into semantic vectors for search, RAG, and clustering.

Overview

Endpoint(s): embeddings-public (curated base) · qwen3-embeddings (Qwen3-Embedding-0.6B)
Model (embeddings-public): sentence-transformers/all-MiniLM-L6-v2
Vector size: 384 dimensions (all-MiniLM-L6-v2)
Capability: text.embeddings
Backend: specialist (runs on CPU or GPU)
Price: per request, USD — see the catalog for the current rate

When to use

✅ Semantic search, retrieval-augmented generation (RAG), document similarity, clustering, recommendations, and duplicate detection.

❌ Very long documents (chunk them first). For cross-account multilingual retrieval consider a multilingual embedding model — import one from HuggingFace.

Request / response

The container accepts a list of texts and returns one vector per text.

from colabhive import ColabHive

client = ColabHive(api_key="hive_...", account_id="YOUR_ACCOUNT_ID")

result = client.endpoints.infer(
    endpoint_id="embeddings-public",   # SDK resolves the name to a UUID
    input_data={"texts": ["cat", "dog", "airplane"]},
)
data = result["result"] if result.get("status") != "queued" \
    else client.endpoints.get_task(result["task_id"])
# data → {"embeddings": [[...384 floats...], ...], "model": "all-MiniLM-L6-v2", "dimensions": 384}

{
  "embeddings": [[0.12, 0.45, ...], [0.23, 0.56, ...]],
  "model": "all-MiniLM-L6-v2",
  "dimensions": 384
}

Field	Type	Description
`texts`	`list[string]`	Texts to embed (batch them for throughput)

The authoritative schema is GET /api/builder/v1/endpoints/{id}.

Use it in a RAG pipeline

# 1) Embed your documents once, store vectors in a vector DB
docs = ["Document 1...", "Document 2...", "Document 3..."]
emb = client.endpoints.infer(endpoint_id="embeddings-public", input_data={"texts": docs})
# 2) Embed the query, retrieve nearest neighbours, then rerank
q = client.endpoints.infer(endpoint_id="embeddings-public", input_data={"texts": ["How do I train a model?"]})
# 3) Improve ordering with the rerank specialist → see text-rerank

Tips

Batch many texts in one call — far more efficient than one call per text.
Cache vectors: embed once, reuse many times.
Cosine similarity on these vectors gives you semantic closeness.

Next steps

Authors: José Luis Minich, Maximiliano Lucius.

Overview​

When to use​

Request / response​

Use it in a RAG pipeline​

Tips​

Next steps​