Skip to main content

Deleting Inferences

There are two primary ways to manage inference storage in Chariot: automated retention policies and manual deletion of specific inferences. This page covers manual deletion options.

When to Use Manual Deletion vs. Retention Policies​

Use Retention Policies for:

  • Automated cleanup based on age
  • Consistent storage management across all inferences
  • Regular housekeeping of old data

Use Manual Deletion for:

  • Removing specific problematic inferences
  • Deleting test data or experimental runs
  • Cleaning up inferences from specific datasets or time periods
  • Immediate removal of sensitive or incorrect data
warning

Deletion is permanent: Once inferences are deleted, both the inference results and associated input data are permanently removed and cannot be recovered.

Delete a Single Inference​

Remove a specific inference and all its associated data:

from chariot.inference_store import inference

deleted_id = inference.delete_inference(model_id=model_id, inference_id=inference_id)
if deleted_id:
    print(f"Successfully deleted inference: {deleted_id}")
else:
    print("Inference not found or deletion failed")

Bulk Delete by Filter​

Delete multiple inferences matching specific criteria. This is useful for cleaning up specific subsets of data.

from chariot.inference_store import models, inference

# Delete all inferences from a specific dataset snapshot
delete_filter = models.BaseInferenceFilter(
    metadata_filter=[
        models.MetadataFilter(
            key="dataset_snapshot_id",
            type=models.MetadataFilterType.STRING,
            operator=models.MetadataFilterOperator.EQUAL,
            value="snapshot-123",
        ),
        models.MetadataFilter(
            key="dataset_split",
            type=models.MetadataFilterType.STRING,
            operator=models.MetadataFilterOperator.EQUAL,
            value="validation",
        ),
    ]
)

deleted_inference_ids = inference.bulk_delete_inferences(
    model_id=model_id, 
    filter_=delete_filter
)

print(f"Deleted {len(deleted_inference_ids)} inferences")