How to add a pipeline to 🤗 Transformers?¶

First and foremost, you need to decide the raw entries the pipeline will be able to take. It can be strings, raw bytes, dictionnaries or whatever seems to be the most likely desired input. Try to keep these inputs as pure Python as possible as it makes compatibility easier (even through other languages via JSON). Those will be the inputs of the pipeline (preprocess).

Then define the outputs. Same policy as the inputs. The simpler, the better. Those will be the outputs of postprocess method.

Start by inheriting the base class Pipeline. with the 4 methods needed to implement preprocess, _forward, postprocess and _sanitize_parameters.

from transformers import Pipeline

class MyPipeline(Pipeline):
    def _sanitize_parameters(self, **kwargs)
        preprocess_kwargs = {}
        if "maybe_arg" in kwargs:
            preprocess_kwargs["maybe_arg"] = kwargs["maybe_arg"]
        return preprocess_kwargs, {}, {}

    def preprocess(self, inputs, maybe_arg=2)
        model_input = Tensor(....)
        return {"model_input": model_input}

    def _forward(self, model_inputs)
        # model_inputs == {"model_input": model_input}
        oututs = self.model(**model_inputs)
        # Maybe {"logits": Tensor(...)}
        return outputs

    def postprocess(self, model_outputs)
        best_class = model_outputs["logits"].softmax(-1)
        return best_class

The structure of this breakdown is to support relatively seemless support for CPU/GPU, while supporting doing pre/postprocessing on the CPU on different threads

preprocess will take the original defined inputs, and turn them something feedable to the model. It might contain more information and is usally a Dict.

_forward is the implementation detail and is not meant to be called directly forward is the preferred called method as it contains safeguards to make sure everything is working on the expected device. If anything is linked to a real model it belongs in the _forward method, anything else is in the preprocess/postrocess.

postprocess methods will take the output of _forward and turn it into the final output that were decided earlier.

_sanitize_parameters exists to allow users to pass any parameters whenever they wish, be it at initialization time pipeline(...., maybe_arg=4) or at call time pipe = pipeline(...); output = pipe(...., maybe_arg=4).

The returns of _sanitize_parameters are the 3 dicts of kwargs that will be passed directly to preprocess, _forward and postprocess. Don’t fill anything if the caller didn’t call with any extra parameter. That allows to keep the default arguments in the function definition which is always more “natural”.

A classic example would be a top_k argument in the post processing in classification tasks.

>>> pipe = pipeline("my-new-task")
>>> pipe("This is a test")
[{"label": "1-star", "score": 0.8}, {"label": "2-star", "score": 0.1}, {"label": "3-star", "score": 0.05}
{"label": "4-star", "score": 0.025}, {"label": "5-star", "score": 0.025}]

>>> pipe("This is a test", top_k=2)
[{"label": "1-star", "score": 0.8}, {"label": "2-star", "score": 0.1}]

In order to achieve that, we’ll update our postprocess method with a default parameter to 5. and edit _sanitize_parameters to allow this new parameter.

def postprocess(self, model_outputs, top_k=5)
    best_class = model_outputs["logits"].softmax(-1)
    # Add logic to handle top_k
    return best_class

def _sanitize_parameters(self, **kwargs)
    preprocess_kwargs = {}
    if "maybe_arg" in kwargs:
        preprocess_kwargs["maybe_arg"] = kwargs["maybe_arg"]

    postprocess_kwargs = {}
    if "top_k" in kwargs:
        preprocess_kwargs["top_k"] = kwargs["top_k"]
    return preprocess_kwargs, {}, postprocess_kwargs

Try to keep the inputs/outputs very simple and ideally JSON-serializable as it makes the pipeline usage very easy without requiring users to understand new kind of objects. It’s also relatively common to support many different types of arguments for ease of use (audio files, can be filenames, URLs or pure bytes)

Adding it to the list of supported tasks¶

Go to src/transformers/pipelines/__init__.py and fill in SUPPORTED_TASKS with your newly created pipeline. If possible it should provide a default model.

Adding tests¶

Create a new file tests/test_pipelines_MY_PIPELINE.py with example with the other tests.

The run_pipeline_test function will be very generic and run on small random models on every possible architecture as defined by model_mapping and tf_model_mapping.

This is very important to test future compatibilty, meaning if someone adds a new model for XXXForQuestionAnswering then the pipeline test will attempt to run on it. Because the models are random it’s impossible to check for actual values, that’s why There is a helper ANY that will simply attempt to match the output of the pipeline TYPE.

You also need to implement 2 (ideally 4) tests.

  • test_small_model_pt : Define 1 small model for this pipeline (doesn’t matter if the results don’t make sense) and test the pipeline outputs. The results should be the same as test_small_model_tf.

  • test_small_model_tf : Define 1 small model for this pipeline (doesn’t matter if the results don’t make sense) and test the pipeline outputs. The results should be the same as test_small_model_pt.

  • test_large_model_pt (optional): Tests the pipeline on a real pipeline where the results are supposed to make sense. These tests are slow and should be marked as such. Here the goal is to showcase the pipeline and to make sure there is no drift in future releases

  • test_large_model_tf (optional): Tests the pipeline on a real pipeline where the results are supposed to make sense. These tests are slow and should be marked as such. Here the goal is to showcase the pipeline and to make sure there is no drift in future releases