Skip to content

replicate

pixeltable.functions.replicate

Pixeltable UDFs that wrap various endpoints from the Replicate API. In order to use them, you must first pip install replicate and configure your Replicate credentials, as described in the Working with Replicate tutorial.

run async 

run(input: Json, *, ref: String) -> Json

Run a model on Replicate.

For additional details, see: https://replicate.com/docs/topics/models/run-a-model

Request throttling: Applies the rate limit set in the config (section replicate, key rate_limit). If no rate limit is configured, uses a default of 600 RPM.

Requirements:

  • pip install replicate

Parameters:

  • input (Json) –

    The input parameters for the model.

  • ref (String) –

    The name of the model to run.

Returns:

  • Json –

    The output of the model.

Examples:

Add a computed column that applies the model meta/meta-llama-3-8b-instruct to an existing Pixeltable column tbl.prompt of the table tbl:

>>> input = {'system_prompt': 'You are a helpful assistant.', 'prompt': tbl.prompt}
... tbl.add_computed_column(response=run(input, ref='meta/meta-llama-3-8b-instruct'))

Add a computed column that uses the model black-forest-labs/flux-schnell to generate images from an existing Pixeltable column tbl.prompt:

>>> input = {'prompt': tbl.prompt, 'go_fast': True, 'megapixels': '1'}
... tbl.add_computed_column(response=run(input, ref='black-forest-labs/flux-schnell'))
... tbl.add_computed_column(image=tbl.response.output[0].astype(pxt.Image))