SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Custom pipeline (#18079)

Browse Source 
* Initial work

* More work

* Add tests for custom pipelines on the Hub

* Protect import

* Make the test work for TF as well

* Last PyTorch specific bit

* Add documentation

* Style

* Title in toc

* Bad names!

* Update docs/source/en/add_new_pipeline.mdx

Co-authored-by: Lysandre Debut <lysandre.debut@reseau.eseo.fr>

* Auto stash before merge of "custom_pipeline" and "origin/custom_pipeline"

* Address review comments

* Address more review comments

* Update src/transformers/pipelines/__init__.py

Co-authored-by: Lysandre Debut <lysandre.debut@reseau.eseo.fr>

Co-authored-by: Lysandre Debut <lysandre.debut@reseau.eseo.fr>
This commit is contained in:
Sylvain Gugger
2022-07-19 12:02:35 +02:00
committed by GitHub
parent 3bb6356d4d
commit dc9147ff36
6 changed files with 439 additions and 57 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
docs/source/en/_toctree.yml
View File

@@ -102,7 +102,7 @@
- local: add_new_model
title: How to add a model to 🤗 Transformers?
- local: add_new_pipeline
title: How to add a pipeline to 🤗 Transformers?
title: How to create a custom pipeline?
- local: testing
title: Testing
- local: pr_checks

139
docs/source/en/add_new_pipeline.mdx
View File

@@ -9,7 +9,10 @@ Unless required by applicable law or agreed to in writing, software distributed
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-->
# How to add a pipeline to 🤗 Transformers?
# How to create a custom pipeline?
In this guide, we will see how to create a custom pipeline and share it on the [Hub](hf.co/models) or add it to the
Transformers library.
First and foremost, you need to decide the raw entries the pipeline will be able to take. It can be strings, raw bytes,
dictionaries or whatever seems to be the most likely desired input. Try to keep these inputs as pure Python as possible
@@ -111,39 +114,123 @@ of arguments for ease of use (audio files, can be filenames, URLs or pure bytes)
## Adding it to the list of supported tasks
To register your `new-task` to the list of supported tasks, provide the
following task template:
```python
my_new_task = {
"impl": MyPipeline,
"tf": (),
"pt": (AutoModelForAudioClassification,) if is_torch_available() else (),
"default": {"model": {"pt": "user/awesome_model"}},
"type": "audio", # current support type: text, audio, image, multimodal
}
```
<Tip>
Take a look at the `src/transformers/pipelines/__init__.py` and the dictionary `SUPPORTED_TASKS` to see how a task is defined.
If possible your custom task should provide a default model.
</Tip>
Then add your custom task to the list of supported tasks via
`PIPELINE_REGISTRY.register_pipeline()`:
To register your `new-task` to the list of supported tasks, you have to add it to the `PIPELINE_REGISTRY`:
```python
from transformers.pipelines import PIPELINE_REGISTRY
PIPELINE_REGISTRY.register_pipeline("new-task", my_new_task)
PIPELINE_REGISTRY.register_pipeline(
"new-task",
pipeline_class=MyPipeline,
pt_model=AutoModelForSequenceClassification,
)
```
You can specify a default model if you want, in which case it should come with a specific revision (which can be the name of a branch or a commit hash, here we took `"abcdef"`) as well was the type:
## Adding tests
```python
PIPELINE_REGISTRY.register_pipeline(
"new-task",
pipeline_class=MyPipeline,
pt_model=AutoModelForSequenceClassification,
default={"pt": ("user/awesome_model", "abcdef")},
type="text", # current support type: text, audio, image, multimodal
)
```
Create a new file `tests/test_pipelines_MY_PIPELINE.py` with example with the other tests.
## Share your pipeline on the Hub
To share your custom pipeline on the Hub, you just have to save the custom code of your `Pipeline` subclass in a
python file. For instance, let's say we want to use a custom pipeline for sentence pair classification like this:
```py
import numpy as np
from transformers import Pipeline
def softmax(outputs):
maxes = np.max(outputs, axis=-1, keepdims=True)
shifted_exp = np.exp(outputs - maxes)
return shifted_exp / shifted_exp.sum(axis=-1, keepdims=True)
class PairClassificationPipeline(Pipeline):
def _sanitize_parameters(self, **kwargs):
preprocess_kwargs = {}
if "second_text" in kwargs:
preprocess_kwargs["second_text"] = kwargs["second_text"]
return preprocess_kwargs, {}, {}
def preprocess(self, text, second_text=None):
return self.tokenizer(text, text_pair=second_text, return_tensors=self.framework)
def _forward(self, model_inputs):
return self.model(**model_inputs)
def postprocess(self, model_outputs):
logits = model_outputs.logits[0].numpy()
probabilities = softmax(logits)
best_class = np.argmax(probabilities)
label = self.model.config.id2label[best_class]
score = probabilities[best_class].item()
logits = logits.tolist()
return {"label": label, "score": score, "logits": logits}
```
The implementation is framework agnostic, and will work for PyTorch and TensorFlow models. If we have saved this in
a file named `pair_classification.py`, we can then import it and register it like this:
```py
from pair_classification import PairClassificationPipeline
from transformers.pipelines import PIPELINE_REGISTRY
from transformers import AutoModelForSequenceClassification, TFAutoModelForSequenceClassification
PIPELINE_REGISTRY.register_pipeline(
"pair-classification",
pipeline_class=PairClassificationPipeline,
pt_model=AutoModelForSequenceClassification,
tf_model=TFAutoModelForSequenceClassification,
)
```
Once this is done, we can use it with a pretrained model. For instance `sgugger/finetuned-bert-mrpc` has been
fine-tuned on the MRPC dataset, which classifies pairs of sentences as paraphrases or not.
```py
from transformers import pipeline
classifier = pipeline("pair-classification", model="sgugger/finetuned-bert-mrpc")
```
Then we can share it on the Hub by using the `save_pretrained` method in a `Repository`:
```py
from huggingface_hub import Repository
repo = Repository("test-dynamic-pipeline", clone_from="{your_username}/test-dynamic-pipeline")
classifier.save_pretrained("test-dynamic-pipeline")
repo.push_to_hub()
```
This will copy the file where you defined `PairClassificationPipeline` inside the folder `"test-dynamic-pipeline"`,
along with saving the model and tokenizer of the pipeline, before pushing everything in the repository
`{your_username}/test-dynamic-pipeline`. After that anyone can use it as long as they provide the option
`trust_remote_code=True`:
```py
from transformers import pipeline
classifier = pipeline(model="{your_username}/test-dynamic-pipeline", trust_remote_code=True)
```
## Add the pipeline to Transformers
If you want to contribute your pipeline to Transformers, you will need to add a new module in the `pipelines` submodule
with the code of your pipeline, then add it in the list of tasks defined in `pipelines/__init__.py`.
Then you will need to add 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`.