Knowledge base quotas and limits

更新时间:
复制 MD 格式

Usage quotas

Category

Description

Limit

Number of knowledge bases

The number of knowledge bases that can be created by each Alibaba Cloud account.

  • Using RDS data source: 100

  • Other data sources: Unlimited

Knowledge base storage capacity (platform storage)

The free vector storage space dedicated to each knowledge base.

  • Ultimate Edition: 9,999 GB

  • Standard Edition: 100 GB

Number of categories

The number of categories (parent + child) that can be created in each workspace.

500

Number of files

The number of files that can be uploaded to each workspace.

100,000

Number of files per knowledge base

The maximum number of files in a single knowledge base.

  • Unstructured knowledge base (for document search): No hard limit

  • Structured knowledge base (for data query): 1 file (a single Excel or CSV file)

Number of data tables

The number of data tables that can be created in each workspace.

1,000

Table limits for AnalyticDB for PostgreSQL (ADB-PG) vector storage

The maximum number of rows and maximum size per row for a single data table when using AnalyticDB for PostgreSQL (ADB-PG) as the vector storage.

  • Maximum rows per table: 10,000,000

  • Maximum size per row: 100 KB

File upload

Supported formats

Knowledge bases for document search

Supported format

Limit

pdf, docx, doc, wps, pptx, ppt

  • Maximum 150 MB

  • Maximum 1,000 pages

txt, markdown, html

Maximum 10 MB

xlsx, xls

Maximum 10 MB, within 100,000 rows

png, jpg / jpeg, bmp, gif

  • Maximum 20 MB

  • Short side > 15 pixels, long side < 8,192 pixels, aspect ratio < 50

Knowledge bases for data query and image Q&A

Supported format

Limit

xlsx, xls

  • Maximum 100,000 rows

  • Maximum 100 columns

Knowledge bases for audio and video search

Supported format

Limit

aac, amr, flac, flv, m4a, mp3, mpeg, ogg, opus, wav, webm, wma, mp4, mkv, avi, mov, wmv

Maximum 512 MB

Upload operations

Category

Description

Limit

Number of files per import

The number of files that can be imported in a single operation in the console.

This limit does not apply to batch imports that use an API, for which a batch size of up to 10,000 files is recommended.

50

Number of tags

The number of tags that can be attached to a single file.

32

Data processing

Segmentation

Category

Description

Limit

Number of text segments

The number of text segments for a single file.

Unlimited

Text segment length

The number of tokens in a single text segment.

6,000

Chunk operation limits

Operations on chunks vary by knowledge base type:

Knowledge base type

Edit chunk

Add chunk

Delete chunk

Document search

Supported

Supported

Supported

Data query

Supported

Supported

Supported

Image Q&A

Supported

Supported

Supported

Audio and video search

Supported

Not supported

Supported

Note

When editing a chunk (UpdateChunk), the content length must be between 10 and 6,000 characters. You can delete up to 10 chunks per request (DeleteChunk).

Vectorization

See Text and multimodal embedding.

Category

Description

Vector model

  • For knowledge bases used for document search, data query, and audio/video search: Supports the text-embedding-v4 and text-embedding-v3 models.

  • For knowledge bases used for image Q&A: Only the multimodal-embedding-v1 model is currently supported.

Vector dimensions

  • text-embedding-v4: 512 dimensions

  • text-embedding-v3: 512 dimensions

  • multimodal-embedding-v1: 1024 dimensions

The preceding vector dimensions cannot be changed.

Retrieval

Category

Description

Limit

Concurrent retrieval

The core retrieval performance of the knowledge base itself. This does not include dependent links, such as calls to a sorting model.

When you create a knowledge base (Ultimate Edition), you can preset a processing policy for dependency chain rate limiting.
  • Ultimate Edition: 50–10,000 queries per second (QPS) (adjustable, corresponding to 1–200 Retrieval Compute Units (RCUs))

  • Standard Edition: 1 QPS (fixed)

Number of retrieved text chunks

The number of text segments that can be retrieved in a single query.

20