This organization hosts the reference packs used by quantgap to measure how far a quantized model has drifted from its original weights.
A pack contains the original model's responses and top next-token probabilities for a fixed probe set. It lets anyone score a quantized model without downloading or running the original model themselves.
Install quantgap and score a compatible quantized model:
pip install "quantgap[llamacpp]" \
--extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cpu
quantgap score hf://<publisher>/<model>-GGUF/<file>.gguf