Search
Rank documents by semantic relevance to a natural-language query.
AI & ML · Const
embeddings
Embeddings convert text into vectors that preserve aspects of its content or meaning. Comparing these vectors enables semantic search, clustering, recommendations, anomaly detection, diversity measurement, and classification.
How it works
The embeddings API transforms input text into a numeric vector intended to preserve aspects of the text's content or meaning. Applications compare these vectors to support semantic similarity tasks such as search, clustering, recommendations, anomaly detection, and classification.
Syntax
client.embeddings.create(input=[text], model="text-embedding-3-small")Parameters
| Name | Type | Description |
|---|---|---|
input | string or array of strings | The text string or strings to convert into embeddings. |
model | string | The name of the embedding model, such as text-embedding-3-small. |
Returns
An embedding vector for each supplied input text.
Key points
- Text embeddings quantify how related text strings are.
- Typical applications include semantic search, clustering, recommendations, anomaly detection, diversity measurement, and classification.
- The request requires input text and an embedding model name.
- Similarity between vectors is useful as a proxy for similarity between the represented data.
- Text embedding models accept strings and return vectors.
Common mistakes
- Keep each input within the selected model’s maximum context length; embedding models do not produce output tokens.
Example: Create a text embedding
from openai import OpenAI
client = OpenAI()
text = "Semantic search finds related meaning."
result = client.embeddings.create(
model="text-embedding-3-small",
input=text,
)
vector = result.data[0].embedding
print(len(vector))Creates an embedding vector for a text string and prints its dimensionality.
When to use it
Clustering
Group texts or items according to vector similarity.
Recommendations
Suggest items whose text representations are related.
Anomaly detection
Identify items with unusually low relatedness to the rest of a collection.
Interview questions
QWhat is an embedding?
An embedding is a vector representation of data that aims to preserve aspects of its content or meaning. Related data tends to produce vectors that are closer together than unrelated data.
Follow-up: How would you use embeddings for retrieval? Generate vectors for documents and queries, then rank documents by vector similarity.
QHow do you obtain an embedding?
Send the text string and embedding model name to the embeddings endpoint. The response contains the resulting embedding vector for the input.
Follow-up: What should be persisted for later search? Store document vectors with their source identifiers and metadata, then embed incoming queries with the same model.
QWhat problems are embeddings useful for?
Common applications include search, clustering, recommendations, anomaly detection, diversity measurement, and classification. These tasks use relationships between vectors to find similar or unusual text.
Follow-up: Why is chunking useful for document search? It lets retrieval compare a query with focused portions of a document instead of one vector representing the entire document.
Related
text-embedding-3-smallembeddings APIsemantic searchclusteringrecommendationsanomaly detectionclassification