Semantic Deduplication UV Script

Part of uv-scripts — self-contained UV scripts you run locally or on Hugging Face Jobs in one command.

Remove duplicate / near-duplicate text samples from a Hugging Face dataset by semantic similarity — clean training data and prevent train/test leakage. Uses SemHash with Model2Vec embeddings: CPU-optimized, no GPU required.

Quick start

# CPU is enough — run on Hugging Face Jobs
hf jobs uv run --flavor cpu-upgrade --secrets HF_TOKEN \
    https://huggingface.co/datasets/uv-scripts/deduplication/raw/main/semantic-dedupe.py \
    your-input-dataset text your-output-dataset

# or locally
uv run https://huggingface.co/datasets/uv-scripts/deduplication/raw/main/semantic-dedupe.py \
    your-input-dataset text your-output-dataset

Arguments are INPUT_DATASET TEXT_COLUMN OUTPUT_DATASET. Tune with --threshold (similarity cutoff) and --max-samples (for testing); run with --help for all options.