snowflake.snowpark.functions.ai_complete

snowflake.snowpark.functions.ai_complete(model: str, prompt: ColumnOrLiteralStr, model_parameters: Optional[dict] = None, response_format: Optional[dict] = None, show_details: Optional[bool] = None, _emit_ast: bool = True) → Column[source]
snowflake.snowpark.functions.ai_complete(model: str, prompt: str, file: Column, model_parameters: Optional[dict] = None, response_format: Optional[dict] = None, show_details: Optional[bool] = None, _emit_ast: bool = True) → Column

Generates a response (completion) for a text prompt using a supported language model.

This function supports three main usage patterns: 1. String prompt only: AI_COMPLETE(model, prompt, …) 2. Prompt with file: AI_COMPLETE(model, prompt, file, …) 3. Prompt object: AI_COMPLETE(model, prompt_object, …)

Parameters:

  • model – A string specifying the model to be used. Different input types have different supported models. See details in AI_COMPLETE.

  • prompt – A string prompt, or a Column object from prompt().

  • file – The FILE type column representing an image. When provided, prompt should be a string.

  • model_parameters –

    Optional object containing model hyperparameters:

    • temperature: A value from 0 to 1 controlling randomness (default: 0)

    • top_p: A value from 0 to 1 controlling diversity (default: 0)

    • max_tokens: Maximum number of output tokens (default: 4096, max: 8192)

    • guardrails: Enable Cortex Guard filtering (default: FALSE)

  • response_format – Optional JSON schema that the response should follow for structured outputs.

  • show_details – Optional boolean flag to return detailed inference information (default: FALSE).

Returns:

response_format and show_details are only supported for single string. When show_details is FALSE and response_format is not specified: Returns a string containing the response. When show_details is FALSE and response_format is specified: Returns an object following the provided format. When show_details is TRUE: Returns a JSON object with detailed inference information including choices, created timestamp, model name, and token usage statistics. See details in AI_COMPLETE Returns.

Examples:

>>> # Basic completion with string prompt
>>> df = session.range(1).select(
...     ai_complete('snowflake-arctic', 'What are large language models?').alias("response")
... )
>>> len(df.collect()[0][0]) > 10
True

>>> # Using model parameters
>>> df = session.range(1).select(
...     ai_complete(
...         model='llama2-70b-chat',
...         prompt='How does a snowflake get its unique pattern?',
...         model_parameters={
...             'temperature': 0.7,
...             'max_tokens': 100
...         }
...     ).alias("response")
... )
>>> result = df.collect()[0][0]
>>> len(result) > 0
True

>>> # With detailed output
>>> df = session.range(1).select(
...     ai_complete(
...         model='mistral-large',
...         prompt='Explain AI in one sentence.',
...         show_details=True
...     ).alias("detailed_response")
... )
>>> result = df.collect()[0][0]
>>> 'choices' in result and 'usage' in result
True

>>> # Prompt with image processing
>>> _ = session.sql("CREATE OR REPLACE TEMP STAGE mystage ENCRYPTION = (TYPE = 'SNOWFLAKE_SSE')").collect()
>>> _ = session.file.put("tests/resources/kitchen.png", "@mystage", auto_compress=False)
>>> df = session.range(1).select(
...     ai_complete(
...         model='claude-4-sonnet',
...         prompt='Extract the kitchen appliances identified in this image. Respond in JSON only with the identified appliances.',
...         file=to_file('@mystage/kitchen.png'),
...     )
... )
>>> result = df.collect()[0][0]
>>> "microwave" in result and "refrigerator" in result
True

>>> # Structured output with response format
>>> response_schema = {
...     'type': 'json',
...     'schema': {
...         'type': 'object',
...         'properties': {
...             'sentiment': {'type': 'string'},
...             'confidence': {'type': 'number'}
...         },
...         'required': ['sentiment', 'confidence']
...     }
... }
>>> df = session.range(1).select(
...     ai_complete(
...         model='llama2-70b-chat',
...         prompt='Analyze the sentiment of this text: I love this product!',
...         response_format=response_schema
...     ).alias("structured_result")
... )
>>> result = df.collect()[0][0]
>>> 'sentiment' in result and 'confidence' in result
True

>>> # Using prompt object from prompt() function
>>> df = session.range(1).select(
...     ai_complete(
...         model='claude-3-7-sonnet',
...         prompt=prompt("Extract the kitchen appliances identified in this image. Respond in JSON only with the identified appliances? {0}", to_file('@mystage/kitchen.png')),
...     )
... )
>>> result = df.collect()[0][0]
>>> "microwave" in result and "refrigerator" in result
True
Copy
言語: 日本語