Import from Hugging Face

How to import datasets from HuggingFace Datasets to HoneyHive.

Since HoneyHive’s datasets don’t follow a fixed schema format, we have an automatic integration with HuggingFace datasets (or any kind of dataset management tool) to import datasets into HoneyHive.

Upload a dataset through the SDK

On a high level, all we need to do is

• define our mapping of inputs-outputs
• importing batch size to setup the integration.

We recommend importing the data in batches of 100 rows at a time.

Prerequisites

• You have already created a project in HoneyHive, as explained here.
• You have an API key for your project, as explained here.

Expected time: few minutes

1. Installation

   To install our SDK, run the following commands in the shell.

   Python

   pip install honeyhive datasets
   
2. Authentication & Imports

   To authenticate your SDK, you need to pass your API key.

   Python

   import honeyhive
   from honeyhive.models import components, operations
   from datasets import load_dataset
   
   hhai = honeyhive.HoneyHive(
     bearer_auth='YOUR_API_KEY',
     server_url='HONEYHIVE_SERVER_URL' # Optional / Required for self-hosted or dedicated deployments
   )
   
3. Create the HoneyHive dataset

   Give your new dataset a name and pass the project name to which you want to associate the dataset.

   Keep the generated dataset_id handy for future reference.

   Python

   eval_dataset = hhai.datasets.create_dataset(request=components.CreateDatasetRequest(
     project='YOUR_PROJECT_NAME',
     name='DATASET_NAME',
   ))
   
   dataset_id = eval_dataset.object.result.inserted_id
   
4. Pass your data in batches with a mapping

   Now, using the dataset_id, you can pass your data list and provide a mapping to the fields.

   We’ll create unique datapoints for each entry in the JSON list. The datapoint_id on those entries will be used for joining traces in experiment runs in the future.

   Any field not defined in the mapping is set on the metadata of the datapoint.

   Python

   dataset = load_dataset("lhoestq/demo1") 
   dataset = list(dataset['train'])  # turn the dataset into a list of dictionaries
   datapoint_ids = []
   
   for i in range(0, len(dataset), 100):
       dataset_request = operations.AddDatapointsRequestBody(
           project = 'YOUR_PROJECT_NAME',
           data = dataset[i:i+100], # list of dictionaries
           mapping = operations.Mapping(
               inputs=[
                 'review', # input fields
               ],
               ground_truth=[],
               history=[]
           ),
       )
   
       datapoints = hhai.datasets.add_datapoints(
           dataset_id = dataset_id, # dataset_id from the previous step
           request_body = dataset_request
       )
   
       datapoint_ids.append(datapoints.object.datapoint_ids)
   

You have successfully uploaded your HuggingFace dataset to HoneyHive using the SDK.

You can now view your dataset in the HoneyHive UI.

Next steps

Running experiments

Learn how to run experiments on your dataset.