Model Provenance Kit: Deep-Signal Weight Fingerprints for Transformer Models

Model Provenance Kit distributes pre-computed deep-signal weight fingerprints for transformer models. The dataset consists of dense numerical features extracted from publicly released model weights on the Hugging Face Hub. It was authored by cisco-ai and last updated on April 30, 2026.

Use Cases

• Model attribution based on extracted weight-level evidence.
• Detecting model tampering or provenance anomalies based on multi-signal fingerprints.
• Comparing and clustering transformer model variants based on their weight features.

Strengths

• Features are pre-computed, saving computational effort for users.
• Data is sourced from publicly released models on the Hugging Face Hub.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

cisco-ai

Collection Method

Features extracted from publicly released transformer model weights on the Hugging Face Hub.

Freshness

Last updated 2026-04-30 00:30:28.