Embeddings - Gwenlake Atlas

embeddings.create returns a typed EmbeddingResponse. Long lists are sent in batches of 100 for you, and index follows the order of the input.

texts = [
    "Do not watch this movie!",
    "This movie is so funny. The acting is wonderful.",
    "Everything about this movie is horrible, from the acting to the editing.",
]

response = client.embeddings.create(input=texts, model="e5-base-v2")
for item in response.data:
    print(item.index, len(item.embedding))

print(response.usage.total_tokens)

A single string is accepted as input too.

Into a DataFrame

import pandas as pd

vectors = pd.DataFrame([item.embedding for item in response.data])

Choosing a model

client.models.list() returns every inference model you can call, embedding models included. Model ids are stable; the catalog of available models evolves with the platform.