Skip to content

Functions

Jump to bottom
Kibae Shin edited this page Sep 17, 2023 · 10 revisions

ONNX Model Functions

pg_onnx_import_model(TEXT, TEXT, BYTEA, JSONB, TEXT)

  • Import the ONNX file.
  • If you want to enable the use of CUDA, use {"cuda": true} or {"cuda": {"device_id": 0}} to specify the device ID.
  • Parameters
    • name(TEXT): Model name
    • version(TEXT): Model version
    • model(BYTEA): ONNX model binary data
    • option(JSONB): Options
    • description(TEXT): Model description
  • Returns
    • boolean: true if the model is successfully imported, false otherwise.
SELECT pg_onnx_import_model(
        'sample_model',
        'v20230101',
        PG_READ_BINARY_FILE('/your_model_path/model.onnx')::bytea,
        '{"cuda": true}'::jsonb,
        'sample model'
    );
 pg_onnx_import_model 
----------------------
 t
(1 row)

pg_onnx_drop_model(TEXT, TEXT)

  • Drop imported model.
  • Parameters
    • name(TEXT): Model name
    • version(TEXT): Model version
  • Returns
    • boolean: true if the model is successfully dropped, false otherwise.
SELECT pg_onnx_drop_model('sample_model', 'v20230101');
 pg_onnx_drop_model 
--------------------
 t

pg_onnx_list_model()

  • List all imported models.
  • Returns list of models(SETOF RECORD)
    • name(TEXT): Model name
    • version(TEXT): Model version
    • option(JSONB): Model options
    • inputs(JSONB): Model input names with type and shape
    • outputs(JSONB): Model output names with type and shape
    • description(TEXT): Model description
    • created_at(TIMESTAMPTZ): Model import time
SELECT *
FROM pg_onnx_list_model();
     name     | version |       option       |                               inputs                               |           outputs           | description    |          created_at
--------------+---------+--------------------+--------------------------------------------------------------------+-----------------------------+---------------
 sample_model | v1      | {"cuda": true}     | {"x": "float32[-1,1]", "y": "float32[-1,1]", "z": "float32[-1,1]"} | {"output": "float32[-1,1]"} | sample model1  | 2023-09-16 03:38:55.088838
 sample_model | v2      | {}                 | {"input_ids": "int64[-1,512]", "onnx::Equal_1": "int64[-1,512]"}   | {"output": "int64[-1]"}     | sample model2  | 2023-09-16 03:38:55.088838
(2 rows)

pg_onnx_inspect_model_bin(BYTEA)

  • Get the type and shape information of the inputs and outputs of the ONNX file.
  • Returns
    • inputs(JSONB): Input type and shape
    • outputs(JSONB): Output type and shape
SELECT *
FROM pg_onnx_inspect_model_bin(
        PG_READ_BINARY_FILE('/your_model_path/model.onnx')::bytea
    );
                               inputs                               |           outputs           
--------------------------------------------------------------------+-----------------------------
 {"x": "float32[-1,1]", "y": "float32[-1,1]", "z": "float32[-1,1]"} | {"output": "float32[-1,1]"}
(1 row)

ONNX Session Functions

pg_onnx_create_session(TEXT, TEXT)

  • Create a new ONNX session.
  • If the model ONNX file is too huge, you might get an error. In this case, change the value of the temp_file_limit setting in PostgreSQL to -1 (no limit).
  • Parameters
    • name(TEXT): Model name
    • version(TEXT): Model version
  • Returns session information(RECORD)
    • name(TEXT): Model name
    • version(TEXT): Model version
    • created_at(BIGINT): Session creation time
    • last_executed_at(BIGINT): Last execution time
    • execution_count(BIGINT): Number of executions
    • inputs(JSONB): Model input names with type and shape
    • outputs(JSONB): Model output names with type and shape
    • option(JSONB): Model options
SELECT *
FROM pg_onnx_create_session('sample_model', 'v20230101')
    model     |  version  | created_at | last_executed_at | execution_count |                               inputs                               |           outputs           |     option      
--------------+-----------+------------+------------------+-----------------+--------------------------------------------------------------------+-----------------------------+-----------------
 sample_model | v20230101 | 1694803952 |                0 |               0 | {"x": "float32[-1,1]", "y": "float32[-1,1]", "z": "float32[-1,1]"} | {"output": "float32[-1,1]"} | {"cuda": false}
(1 row)

pg_onnx_describe_session(TEXT, TEXT)

  • Gets information about an ONNX session.
  • Parameters
    • name(TEXT): Model name
    • version(TEXT): Model version
  • Returns session information(RECORD)
    • name(TEXT): Model name
    • version(TEXT): Model version
    • created_at(BIGINT): Session creation time
    • last_executed_at(BIGINT): Last execution time
    • execution_count(BIGINT): Number of executions
    • inputs(JSONB): Model input names with type and shape
    • outputs(JSONB): Model output names with type and shape
    • option(JSONB): Model options
SELECT *
FROM pg_onnx_describe_session('sample_model', 'v20230101')
    model     |  version  | created_at | last_executed_at | execution_count |                               inputs                               |           outputs           |     option      
--------------+-----------+------------+------------------+-----------------+--------------------------------------------------------------------+-----------------------------+-----------------
 sample_model | v20230101 | 1694803952 |                0 |               0 | {"x": "float32[-1,1]", "y": "float32[-1,1]", "z": "float32[-1,1]"} | {"output": "float32[-1,1]"} | {"cuda": false}
(1 row)

pg_onnx_execute_session(TEXT, TEXT, JSONB)

  • Execute the ONNX session.
  • ${\color{red}Create an ONNX session if it doesn't exist.}$
  • Parameters
    • name(TEXT): Model name
    • version(TEXT): Model version
    • inputs(JSONB): Input data
  • Returns
    • outputs(JSONB): Output data
SELECT pg_onnx_execute_session('sample_model', 'v20230101', '{
  "x": [[1], [2], [3]],
  "y": [[3], [4], [5]],
  "z": [[5], [6], [7]]
}');
                                pg_onnx_execute                                 
--------------------------------------------------------------------------------
 {"output": [[0.7488641738891602], [0.8607008457183838], [0.9725375175476074]]}
(1 row)

pg_onnx_destroy_session(TEXT, TEXT, JSONB)

  • Destroy the ONNX session. Clean up unused ONNX sessions.
  • Parameters
    • name(TEXT): Model name
    • version(TEXT): Model version
  • Returns
    • boolean: true if the model is successfully destroyed, false otherwise.
SELECT pg_onnx_destroy_session('sample_model', 'v20230101');
 pg_onnx_destroy_session 
-------------------------
 t
(1 row)

pg_onnx_list_session()

  • List all active ONNX sessions.
  • Returns list of sessions(SETOF RECORD)
    • name(TEXT): Model name
    • version(TEXT): Model version
    • created_at(BIGINT): Session creation time
    • last_executed_at(BIGINT): Last execution time
    • execution_count(BIGINT): Number of executions
    • inputs(JSONB): Model input names with type and shape
    • outputs(JSONB): Model output names with type and shape
    • option(JSONB): Model options
SELECT *
FROM pg_onnx_list_model();
    model     |  version  | created_at | last_executed_at | execution_count |                               inputs                               |           outputs           |     option      
--------------+-----------+------------+------------------+-----------------+--------------------------------------------------------------------+-----------------------------+-----------------
 sample_model | v20230101 | 1694803952 |                0 |               0 | {"x": "float32[-1,1]", "y": "float32[-1,1]", "z": "float32[-1,1]"} | {"output": "float32[-1,1]"} | {"cuda": false}
(1 row)
Clone this wiki locally