After a cuML cluster step, one cluster is almost all garbled OCR lines from a city tax-assessor's scanned invoices. How should that cluster be used?
Select an answer to reveal the explanation.
Short Explanation
One cluster is almost all garbled OCR from scanned invoices. Drop or quarantine that cluster as a data-quality filter. Cluster purity is not the project score, and TensorRT-LLM will not clean the junk.
Full Explanation
Supervised or unsupervised GPU models can prepare a dataset by finding junk, outliers, or near-duplicates. A cluster of garbled OCR is a data-quality filter, not the reported score. That work stays in Domain 4. Do not treat cluster purity as an experiment success bar or push the junk through an inference optimizer.