Milvus

Batch process all your records to store structured outputs in Milvus.

The requirements are as follows.

• For the Unstructured UI or the Unstructured API, only Milvus cloud-based instances (such as Zilliz Cloud, and Milvus on IBM watsonx.data) are supported.
• For Unstructured Ingest, Milvus local and cloud-based instances are supported.

The following video shows how to fulfill the minimum set of requirements for Milvus cloud-based instances, demonstrating Milvus on IBM watsonx.data:

• For Zilliz Cloud, you will need:

  • A Zilliz Cloud account.

  • A Zilliz Cloud cluster.

  • The URI of the cluster, also known as the cluster’s public endpoint, which takes a format such as https://<cluster-id>.<cluster-type>.<cloud-provider>-<region>.cloud.zilliz.com. Get the cluster’s public endpoint.

  • The token to access the cluster. Get the cluster’s token.

  • The name of the database in the instance.

  • The name of the collection in the database.

    The collection must have a a defined schema before Unstructured can write to the collection. The minimum viable schema for Unstructured contains only the fields element_id, embeddings, and record_id, as follows:

    Field Name	Field Type	Max Length	Dimension	Index	Metric Type
    element_id (primary key field)	VARCHAR	200	—	—	—
    embeddings (vector field)	FLOAT_VECTOR	—	3072	Yes (Checked)	Cosine
    record_id	VARCHAR	200	—	—	—

• For Milvus on IBM watsonx.data, you will need:

  • An IBM Cloud account.
  • The IBM watsonx.data subscription plan.
  • A Milvus service instance in IBM watsonx.data.
  • The URI of the instance, which takes the format of https://, followed by instance’s GRPC host, followed by a colon and the GRPC port. This takes the format of https://<host>:<port>. Get the instance’s GRPC host and GRPC port.
  • The name of the database in the instance.
  • The name of the collection in the database. Note the collection requirements at the end of this section.
  • The uername and password to access the instance. The username for Milvus on IBM watsonx.data is always ibmlhapikey. The password for Milvus on IBM watsonx.data is in the form of an IBM Cloud user API key. Get the user API key.

• For Milvus local, you will need:

  • A Milvus instance.
  • The URI of the instance.
  • The name of the database in the instance.
  • The name of the collection in the database. Note the collection requirements at the end of this section.
  • The username and password, or token to access the instance.

All Milvus instances require the target collection to have a defined schema before Unstructured can write to the collection. The minimum viable schema for Unstructured contains only the fields element_id, embeddings, and record_id, as follows. This example code demonstrates the use of the Python SDK for Milvus to create a collection with this minimum viable schema, targeting Milvus on IBM watsonx.data. For the connections.connect arguments to connect to other types of Milvus deployments, see your Milvus provider’s documentation:

import os
from pymilvus import (
    connections,
    FieldSchema,
    DataType,
    CollectionSchema,
    Collection,
)

connections.connect(
    alias="default",
    host=os.getenv("MILVUS_GRPC_HOST"),
    port=os.getenv("MILVUS_GRPC_PORT"),
    user=os.getenv("MILVUS_USER"),
    password=os.getenv("MILVUS_PASSWORD"),
    secure=True
)

primary_key = FieldSchema(
    name="element_id",
    dtype=DataType.VARCHAR,
    is_primary=True,
    max_length=200
)

vector = FieldSchema(
    name="embeddings",
    dtype=DataType.FLOAT_VECTOR,
    dim=3072
)

record_id = FieldSchema(
    name="record_id",
    dtype=DataType.VARCHAR,
    max_length=200
)

schema = CollectionSchema(
    fields=[primary_key, vector, record_id],
    enable_dynamic_field=True
)

collection = Collection(
    name="my_collection",
    schema=schema,
    using="default"
)

index_params = {
    "metric_type": "L2",
    "index_type": "IVF_FLAT",
    "params": {"nlist": 1024}
}

collection.create_index(
    field_name="embeddings",
    index_params=index_params
)

Other approaches, such as creating collections instantly or setting nullable and default fields, have not been fully evaluated by Unstructured and might produce unexpected results.

Unstructured cannot provide a schema that is guaranteed to work in all circumstances. This is because these schemas will vary based on your source files’ types; how you want Unstructured to partition, chunk, and generate embeddings; any custom post-processing code that you run; and other factors.

The Milvus connector dependencies:

pip install "unstructured-ingest[milvus]"

You might also need to install additional dependencies, depending on your needs. Learn more.

The following environment variables:

• MILVUS_URI - The Milvus instance’s URI, represented by --uri (CLI) or uri (Python).
• MILVUS_USER and MILVUS_PASSWORD, or MILVUS_TOKEN - The username and password, or token, to access the instance. This is represented by --user and --password, or --token (CLI); or user and password, or token (Python).
• MILVUS_DB - The database’s name, represented by --db-name (CLI) or db_name (Python).
• MILVUS_COLLECTION - The collection’s name, represented by --collection-name (CLI) or collection_name (Python).
• MILVUS_FIELDS_TO_INCLUDE - A list of fields to include a comma-separated list (CLI) or an array of strings (Python), represented by --field-to-include (CLI) or fields_to_include (Python).

Additional settings include:

• To emit the metadata field’s child fields directly into the output, include --flatten-metadata (CLI) or flatten_metadata=True (Python). This is the default if not specified.
• To keep the metadata field with its child fields intact in the output, include --no-flatten-metadata (CLI) or flatten_metadata=False (Python).

Now call the Unstructured CLI or Python. The source connector can be any of the ones supported. This example uses the local source connector:

This example sends files to Unstructured for processing by default. To process files locally instead, see the instructions at the end of this page.

#!/usr/bin/env bash

# Chunking and embedding are optional.

unstructured-ingest \
  local \
    --input-path $LOCAL_FILE_INPUT_DIR \
    --chunking-strategy by_title \
    --embedding-provider huggingface \
    --partition-by-api \
    --api-key $UNSTRUCTURED_API_KEY \
    --partition-endpoint $UNSTRUCTURED_API_URL \
    --strategy hi_res \
    --additional-partition-args="{\"split_pdf_page\":\"true\", \"split_pdf_allow_failed\":\"true\", \"split_pdf_concurrency_level\": 15}" \
  milvus \
    --uri $MILVUS_URI \
    --user $MILVUS_USER \
    --password $MILVUS_PASSWORD \
    --db-name $MILVUS_DB \
    --collection-name $MILVUS_COLLECTION \
    --fields-to-include type,element_id,text,embeddings

For the Unstructured Ingest CLI and the Unstructured Ingest Python library, you can use the --partition-by-api option (CLI) or partition_by_api (Python) parameter to specify where files are processed:

• To do local file processing, omit --partition-by-api (CLI) or partition_by_api (Python), or explicitly specify partition_by_api=False (Python).

  Local file processing does not use an Unstructured API key or API URL, so you can also omit the following, if they appear:

  • --api-key $UNSTRUCTURED_API_KEY (CLI) or api_key=os.getenv("UNSTRUCTURED_API_KEY") (Python)
  • --partition-endpoint $UNSTRUCTURED_API_URL (CLI) or partition_endpoint=os.getenv("UNSTRUCTURED_API_URL") (Python)
  • The environment variables UNSTRUCTURED_API_KEY and UNSTRUCTURED_API_URL

• To send files to the Unstructured Partition Endpoint for processing, specify --partition-by-api (CLI) or partition_by_api=True (Python).

  Unstructured also requires an Unstructured API key and API URL, by adding the following:

  • --api-key $UNSTRUCTURED_API_KEY (CLI) or api_key=os.getenv("UNSTRUCTURED_API_KEY") (Python)
  • --partition-endpoint $UNSTRUCTURED_API_URL (CLI) or partition_endpoint=os.getenv("UNSTRUCTURED_API_URL") (Python)
  • The environment variables UNSTRUCTURED_API_KEY and UNSTRUCTURED_API_URL, representing your API key and API URL, respectively.

  Get an API key and API URL.