Install custom python libraries in HuggingFace DLC

Amazon SageMaker
1

Hi,

I have the source code for a custom data cleaning library; the code is private to my company. I usually install it from local source using something ‘pip install ./cleanse’. I know the requirements.txt file can be used to specify package versions for public libraries like pandas and NumPy but how can we install custom packages in the HuggingFace DLC?

2

You can put ./cleanse into you source_dir and then either add os.system("pip install ./cleanse") at the top of your training script or add the local path ./cleanse into a requirements.txt in the source_dir.

3

Hi, I’m actually using this library to clean in inference. Is there a place I could leverage the inference.py file inside the code/ directory in the model.tar.gz file? Or otherwise can I somehow integrate the into the preprocess script (using SKLearnProcessor)

4

Sorry, i am not sure what you mean with Is there a place I could leverage the inference.py file inside the code/ directory in the model.tar.gz file?

5

I’m using an inference.py file right now as per instructions at the bottom of this page to override defaults during model inference. I’m wondering how/if I can use the custom library within this file

6

^like that

7

What is source_dir though? Like where is that argument specified

8

How are you deploying your model currently? could you share the code snippet? source_dir refers to the HuggingFaceModel of the sagemaker-sdk.

9 
# create Hugging Face Model Class
huggingface_model = HuggingFaceModel(
   model_data= model_uri,  # path to your trained sagemaker model
   role=role, # iam role with permissions to create an Endpoint
   transformers_version="4.17", # transformers version used
   pytorch_version="1.10", # pytorch version used
   py_version="py38", 
   env={ 'HF_TASK':'text-classification' }# python version of the DLC
)
10

does the model_uri contain a code/inference.py ? If so you need to put the requirements.txt / ./cleanse into the code/ directory.

1 Like