jina-embeddings-v4
Jina embeddings v4 on Pinecone: multimodal text and image embeddings with 32k tokens, flexible 128-2048 dims, and multi-vector retrieval.
jina-embeddings-v4
Section titled “jina-embeddings-v4”Overview
Section titled “Overview”jina-embeddings-v4 is a universal embedding model for multimodal and multilingual retrieval. The model is specially designed for complex document retrieval, including visually rich documents with charts, tables, and illustrations.
Features
Section titled “Features”jina-embeddings-v4 features:
- Unified embeddings for text, images, and visual documents, supporting both dense (single-vector) and late-interaction (multi-vector) retrieval.
- Multilingual support (30+ languages) and compatibility with a wide range of domains, including technical and visually complex documents.
- Task-specific adapters for retrieval, text matching, and code-related tasks, which can be selected at inference time.
- Flexible embedding size: dense embeddings are 2048 dimensions by default but can be truncated to as low as 128 with minimal performance loss.
Installation
Section titled “Installation”pip install pinecone requestsCreate Index
Section titled “Create Index”from pinecone import Pinecone, ServerlessSpec
pc = Pinecone(api_key="API_KEY")
JINA_API_KEY = ""
dimension = 1024
index_name = "jina-embeddings-v4"
if not pc.has_index(index_name):
pc.create_index(
name=index_name,
dimension=dimension,
metric="cosine",
spec=ServerlessSpec(
cloud='aws',
region='us-east-1'
)
)
index = pc.Index(index_name)Embed & Upsert
Section titled “Embed & Upsert”from typing import Literal, List
import requests
def get_embeddings(
texts: List[str],
dimensions: int,
task: Literal['text-matching', 'separation', 'classification', 'retrieval.query', 'retrieval.passage']):
headers = {
'Content-Type': 'application/json',
'Authorization': f'Bearer {JINA_API_KEY}'
}
data = {
'input': texts,
'model': 'jina-embeddings-v4',
'dimensions': dimensions,
'task': task
}
response = requests.post('https://api.jina.ai/v1/embeddings', headers=headers, json=data)
return response.json()
# Data to index
data = [
{"id": "vec1", "text": "Apple is a popular fruit known for its sweetness and crisp texture."},
{"id": "vec2", "text": "The tech company Apple is known for its innovative products like the iPhone."},
{"id": "vec3", "text": "Many people enjoy eating apples as a healthy snack."},
{"id": "vec4", "text": "Apple Inc. has revolutionized the tech industry with its sleek designs and user-friendly interfaces."},
{"id": "vec5", "text": "An apple a day keeps the doctor away, as the saying goes."},
]
embeddings = get_embeddings([d["text"] for d in data], dimensions=dimension, task='retrieval.passage')
embeddings = [e["embedding"] for e in embeddings["data"]]
vectors = []
for d, e in zip(data, embeddings):
vectors.append({
"id": d['id'],
"values": e,
"metadata": {'text': d['text']}
})
index.upsert(
vectors=vectors,
namespace="ns1"
)query = "Tell me about the tech company known as Apple"
# Remember to keep query and document embedding to the same dimensions
x = get_embeddings([query], dimensions=dimension, task='retrieval.query')["data"][0]["embedding"]
results = index.query(
namespace="ns1",
vector=x,
top_k=3,
include_values=False,
include_metadata=True
)
print(results)Lorem Ipsum