bhavsarpratik / serverless-transformers-on-aws-lambda

Deploy transformers serverless on AWS Lambda

Hey I'm Pratik 👋

A remote NLP engineer from India 👾

NLP | Deep Learning | Python | AWS

I am a self-taught data scientist from India working remotely from Mumbai. I don't know if my skills are wide or deep, but I can get things done 🧗

I have worked on a variety of domains and projects.

• Semantic search using Deep learning
• Elasticsearch - search and recommendation engine
• Text classification(English and Hinglish)
• Chatbots
• Algorithmic trading with ML
• Applying AutoML

🧪 Some of my best projects are available at hackathons

🖌 I regulary write data science blogs on Substack & Modern NLP

🙏🏼 You can reach me via Twitter

🏇 You can find my community talks at YouTube

serverless-transformers-on-aws-lambda's Issues

Dockerbuild fails - "chmod: cannot access ‘./models’: No such file or directory"

Am I missing something here? There is no ./models folder but we are trying to use it for something? @bhavsarpratik

Add text-generation pipeline

Include HF text generation pipeline

Zaero Shot Classification Test Fails While Creating Docker Image

Hey,
I was trying to use Zaero-Shot-Classification Pipeline and build its Docker Image but it fails while running pytests tests command,
I am sharing the error log below, please help me out with this. I have not changed anything in the test cases, it is the same as it is provided in this repository.
.

#11 170.9 __________________________ test_multi_label_response ___________________________
#11 170.9
#11 170.9 requests = ({'hypothesis': 'The sentiment of the review is {}.', 'labels': ['negative', 'postive', 'neutral'], 'model_name': 'typ...m going out for food']}, {'multi_label': True, 'texts': ['food was great', 'food was bad', 'i
am going out for food']})
#11 170.9 response = ({'predictions': [{'label': 'postive', 'score': 0.8}, {'label': 'negative', 'score': 0.87}, {'label': 'postive', 'scor... 'postive'], 'score': [1.0, 0.85, 0.83]}, {'label': ['postive', 'negative', 'neutral'], 'scor
e': [0.67, 0.34, 0.14]}]})
#11 170.9
#11 170.9 def test_multi_label_response(requests, response):
#11 170.9 > assert response[2] == pipeline(requests[2])
#11 170.9
#11 170.9 tests/test_classifier.py:12:
#11 170.9 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
#11 170.9 src/classifier.py:63: in __call__
#11 170.9 predictions = classification_pipeline(texts, labels, hypothesis, multi_label=multi_label)
#11 170.9 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
#11 170.9
#11 170.9 self = <transformers.pipelines.zero_shot_classification.ZeroShotClassificationPipeline object at 0x7f08ca4f9f70>
#11 170.9 sequences = ['food was great', 'food was bad', 'i am going out for food']
#11 170.9 args = (['negative', 'postive', 'neutral'], 'The sentiment of the review is {}.')
#11 170.9 kwargs = {'multi_label': True}
#11 170.9
#11 170.9 def __call__(
#11 170.9 self,
#11 170.9 sequences: Union[str, List[str]],
#11 170.9 *args,
#11 170.9 **kwargs,
#11 170.9 ):
#11 170.9 """
#11 170.9 Classify the sequence(s) given as inputs. See the [`ZeroShotClassificationPipeline`] documentation for more
#11 170.9 information.
#11 170.9
#11 170.9 Args:
#11 170.9 sequences (`str` or `List[str]`):
#11 170.9 The sequence(s) to classify, will be truncated if the model input is too large.
#11 170.9 candidate_labels (`str` or `List[str]`):
#11 170.9 The set of possible class labels to classify each sequence into. Can be a single label, a string of
#11 170.9 comma-separated labels, or a list of labels.
#11 170.9 hypothesis_template (`str`, *optional*, defaults to `"This example is {}."`):
#11 170.9 The template used to turn each label into an NLI-style hypothesis. This template must include a {} or
#11 170.9 similar syntax for the candidate label to be inserted into the template. For example, the default
#11 170.9 template is `"This example is {}."` With the candidate label `"sports"`, this would be fed into the
#11 170.9 model like `"<cls> sequence to classify <sep> This example is sports . <sep>"`. The default template
#11 170.9 works well in many cases, but it may be worthwhile to experiment with different templates depending on
#11 170.9 the task setting.
#11 170.9 multi_label (`bool`, *optional*, defaults to `False`):
#11 170.9 Whether or not multiple candidate labels can be true. If `False`, the scores are normalized such that
#11 170.9 the sum of the label likelihoods for each sequence is 1. If `True`, the labels are considered
#11 170.9 independent and probabilities are normalized for each candidate by doing a softmax of the entailment
#11 170.9 score vs. the contradiction score.
#11 170.9
#11 170.9 Return:
#11 170.9 A `dict` or a list of `dict`: Each result comes as a dictionary with the following keys:
#11 170.9
#11 170.9 - **sequence** (`str`) -- The sequence for which this is the output.
#11 170.9 - **labels** (`List[str]`) -- The labels sorted by order of likelihood.
#11 170.9 - **scores** (`List[float]`) -- The probabilities for each of the labels.
#11 170.9 """
#11 170.9 if len(args) == 0:
#11 170.9 pass
#11 170.9 elif len(args) == 1 and "candidate_labels" not in kwargs:
#11 170.9 kwargs["candidate_labels"] = args[0]
#11 170.9 else:
#11 170.9 > raise ValueError(f"Unable to understand extra arguments {args}")
#11 170.9 E ValueError: Unable to understand extra arguments (['negative', 'postive', 'neutral'], 'The sentiment of the review is {}.')
#11 170.9
#11 170.9 ../lang/lib/python3.9/site-packages/transformers/pipelines/zero_shot_classification.py:179: ValueError
#11 170.9 =========================== short test summary info ============================
#11 170.9 FAILED tests/test_classifier.py::test_complete_response - ValueError: Unable ...
#11 170.9 FAILED tests/test_classifier.py::test_default_response - ValueError: Unable t...
#11 170.9 FAILED tests/test_classifier.py::test_multi_label_response - ValueError: Unab...
#11 170.9 ======================== 3 failed in 160.47s (0:02:40) =========================
#11 ERROR: executor failed running [/bin/sh -c pip install pytest --no-cache-dir && pytest tests -s -vv]: exit code: 1

Remove hardcoded image URI in workflows/classification.yml?

968911158010.dkr.ecr.us-east-1.amazonaws.com ?

Add AWS SAM for creating Lambda and API Gateway

SAM code for pipeline - zero-shot-classification in zero-shot-classification/sam

• create lambda with name same as pipeline directory
• use the pipeline name ECR registry with container tagged as latest
• use configuration of 2000 RAM and 30 sec timeout
• create API gateway
  • add resource name same as pipeline name
  • add post method to the path /pipeline-name
• add this creation step commands to pipeline github workflow (should create lambda and API if they do not exist)

Kindly test it end2end on your AWS account before creating the PR

Add sentence encoding pipeline

Add sentence encoding pipeline using sentence-transformers