Realtime Compute for Apache Flink:Invocação geral

CREATE TEMPORARY TABLE text_source (
  user_input STRING
) WITH ('connector' = 'datagen');

CREATE TEMPORARY TABLE result_sink (
  user_input STRING,
  ai_analysis STRING
) WITH ('connector' = 'print');

CREATE TEMPORARY MODEL text_model
INPUT (user_input STRING)
OUTPUT (content STRING)
WITH (
  'provider' = 'openai-compat',
  'model' = 'qwen3.6-flash',
  'task' = 'chat/completions',
  'system-prompt' = 'Rate the gibberish level of the input on a scale of 0 to 100'
);

INSERT INTO result_sink
SELECT user_input, content as ai_analysis FROM
    ML_PREDICT(
        TABLE text_source,
        MODEL text_model,
        DESCRIPTOR(user_input)
);

ML_PREDICT(TABLE <table_name>, MODEL <model_name>, DESCRIPTOR(<input_columns>) [, CONFIG => MAP[...]])        

Parâmetros de tipo de conteúdo

• Para entradas de coluna única, use content-type para especificar o tipo de conteúdo. Valores suportados: text, image_url.

• Em entradas multicoluna, utilize content-types para definir o tipo de conteúdo de cada coluna. Valores suportados: text, image_url, multi_image_urls.

• Apenas um entre content-type e content-types pode ser especificado. A opção disponível depende do que foi definido no CREATE MODEL:

Parâmetros no nível da coluna

Texto

O exemplo a seguir registra e utiliza um modelo integrado do Flink AI Service para classificar o sentimento de textos de entrada. O Exemplo 1 usa parâmetros no nível do MODEL. O Exemplo 2 substitui o user-prompt no momento da chamada.

-- Register a built-in model. When content-type is not specified, it defaults to text
CREATE MODEL sentiment_model
INPUT (prompt STRING)
OUTPUT (response STRING)
WITH (
  'provider' = 'openai-compat',
  'task' = 'chat/completions',
  'model' = 'qwen3.6-flash',
  'system-prompt' = 'You are a sentiment classifier. Output one label: negative, positive, or neutral'
);

-- Create a source table: simulated product review data
CREATE TEMPORARY VIEW input_table(id, content)
AS VALUES
  (1, 'Great quality, soft fabric, fits perfectly'),
  (2, 'Had loose threads on arrival, faded badly after one wash'),
  (3, 'Received the item, looks as pictured'),
  (4, 'Started pilling after two weeks, customer service refused returns'),
  (5, 'Flattering fit, color is even better than the photo, already ordered a third one');

-- Create a result table
CREATE TEMPORARY TABLE output_table (
  id INT,
  content STRING,
  sentiment STRING
) WITH (
  'connector' = 'print'
);

-- Use ML_PREDICT for real-time inference
-- Example 1: Use MODEL-level parameters
INSERT INTO output_table
SELECT
  id,
  content,
  response AS sentiment
FROM ML_PREDICT(
  TABLE input_table,
  MODEL sentiment_model,
  DESCRIPTOR(content));

-- Example 2: Specify per-call parameters
INSERT INTO output_table
SELECT
  id,
  content,
  response AS sentiment
FROM ML_PREDICT(
  TABLE input_table,
  MODEL sentiment_model,
  DESCRIPTOR(content),
  MAP['user-prompt', 'Reply in Chinese']);

Saída do Exemplo 1:

Saída do Exemplo 2:

Imagem

Este exemplo registra um modelo multimodal e classifica imagens de entrada.

-- Register a built-in model with content-type set to image_url
CREATE TEMPORARY MODEL sentiment_model
INPUT (prompt STRING)
OUTPUT (response STRING)
WITH (
  'provider' = 'openai-compat',
  'task' = 'chat/completions',
  'model' = 'qwen3.6-flash',
  'content-type' = 'image_url'
);

-- Use ML_PREDICT for real-time inference
INSERT INTO output_table
SELECT
  id,
  content,
  response AS sentiment
FROM ML_PREDICT(
  TABLE input_table,
  MODEL sentiment_model,
  DESCRIPTOR(content));

Texto + imagem única

O exemplo abaixo registra um modelo multimodal para inferência baseada em chat com entradas de texto e imagem.

CREATE MODEL vl_model
INPUT (text_input STRING, image_input STRING)
OUTPUT (content STRING)
WITH (
  'provider' = 'openai-compat',
  'model' = 'qwen3.5-plus',
  'task' = 'chat/completions',
  'content-types' = 'text;image_url'
);

INSERT INTO result_sink
SELECT content FROM TABLE(ML_PREDICT(
  TABLE image_source,
  MODEL vl_model,
  DESCRIPTOR(text_input, image_input)
));         

Texto + múltiplas imagens (múltiplas colunas)

Passe cada imagem por meio de uma coluna INPUT separada. Especifique image_url para cada coluna de imagem em content-types.

CREATE MODEL vl_model_multi
INPUT (prompt STRING, img1 STRING, img2 STRING)
OUTPUT (content STRING)
WITH (
  'provider' = 'openai-compat',
  'model' = 'qwen3.5-plus',
  'task' = 'chat/completions',
  'content-types' = 'text;image_url;image_url'
);

SELECT content FROM TABLE(ML_PREDICT(
  TABLE my_source,
  MODEL vl_model_multi,
  DESCRIPTOR(prompt, img1, img2)
));          

Texto + múltiplas imagens (array)

Passe várias imagens em uma coluna ARRAY<STRING>. Use multi_image_urls em content-types.

CREATE MODEL vl_model_array
INPUT (prompt STRING, images ARRAY<STRING>)
OUTPUT (content STRING)
WITH (
  'provider' = 'openai-compat',
  'model' = 'qwen3.5-plus',
  'task' = 'chat/completions',
  'content-types' = 'text;multi_image_urls'
);

SELECT content FROM TABLE(ML_PREDICT(
  TABLE my_source,
  MODEL vl_model_array,
  DESCRIPTOR(prompt, images)
)); 

-- When content-type / content-types are not specified in CREATE MODEL, the default is content-type = text
CREATE MODEL model_single
INPUT (input STRING)
OUTPUT (content STRING)
WITH (
  'provider' = 'openai-compat',
  'model' = 'qwen3.5-plus',
  'task' = 'chat/completions'
);

-- Call 1: Use MODEL parameters, content type is text
SELECT content FROM TABLE(ML_PREDICT(
  TABLE source_text,
  MODEL model_single,
  DESCRIPTOR(input)
));

-- Call 2: Override content type to image at call time
SELECT content FROM TABLE(ML_PREDICT(
  TABLE source_img,
  MODEL model_single,
  DESCRIPTOR(input),
  MAP['content-type', 'image_url']
));

-- Set default configuration in CREATE MODEL (multi-column model)
CREATE MODEL vl_model
INPUT (text_input STRING, image_input STRING)
OUTPUT (content STRING)
WITH (
  'provider' = 'openai-compat',
  'model' = 'qwen3.5-plus',
  'task' = 'chat/completions',
  'content-types' = 'text;image_url',
  'user-prompt' = 'Describe the image'
);

-- Call 1: Use default configuration
SELECT content FROM TABLE(ML_PREDICT(
  TABLE source_a, MODEL vl_model, DESCRIPTOR(text_input, image_input)
));

-- Call 2: Override user-prompt and column-level parameters
SELECT content FROM TABLE(ML_PREDICT(
  TABLE source_b, MODEL vl_model, DESCRIPTOR(text_input, image_input),
  MAP[
    'user-prompt', 'Answer in English',
    'image_input.min_pixels', '100',
    'image_input.max_pixels', '5000'
  ]
));

-- Call 3: Override content-types to change the type combination (treat both columns as text)
SELECT content FROM TABLE(ML_PREDICT(
  TABLE source_c, MODEL vl_model, DESCRIPTOR(text_input, image_input),
  MAP['content-types', 'text;text']
));

-- Call 4: Override content-types to change the type combination (reverse order)
SELECT content FROM TABLE(ML_PREDICT(
  TABLE source_c, MODEL vl_model, DESCRIPTOR(text_input, image_input),
  MAP['content-types', 'image_url;text']
));