Implement `replicate.run()` method for Python client

Question

Implement `replicate.run()` method for Python client

Closed this issue 3 months ago · 3 comments

replicate.run is a high-level abstraction for running models and getting their output.

We want to support replicate.run in the new Stainless client and maintain feature parity with the legacy client, to minimize the migration cost for users upgrading to the new client.

Reference

Notes

Use the predictions.create HTTP API to run "versioned" models.
Use the models.predictions.create HTTP API to run official models
Support wait option
Port tests from old client
Document the method

Answer 1 · 2025-06-30T22:32:06.000Z

@dgellow @dtmeadows where are we with this one?

Answer 2 · 2025-07-31T14:32:33.000Z

tests have been ported
docs haven't been ported yet

Use the predictions.create HTTP API to run "versioned" models

Use the models.predictions.create HTTP API to run official models

I believe that's the case, as seen here: https://github.com/replicate/replicate-python-stainless/blob/bda7c30d833b4b7b30fb4716687ffcbbd95e1cd5/src/replicate/lib/_predictions.py#L131-L141

Answer 3 · 2025-07-31T14:41:01.000Z

I believe we are now all done here!