Weightscope is an open-source command-line tool designed to provide detailed, tensor-level comparisons between AI model weights, with a particular focus on HuggingFace checkpoints. It addresses the challenge of understanding what has actually changed between different versions of models, moving beyond vague changelogs and release notes to generate concrete, measurable diff reports. By enabling users to see which tensors have changed and by how much, Weightscope brings transparency to model updates and fine-tuning processes.
The tool allows users to stream and compare any two HuggingFace model checkpoints directly from the Hub without requiring downloads or setup. It produces a real diff report, highlighting metrics such as the percentage of weights changed, average delta per tensor, and layer-specific changes. Weightscope supports both LoRA and full fine-tune visibility, making it possible to distinguish between minor adjustments and substantial retraining. The output is delivered as a markdown report, suitable for inclusion in pull requests, model cards, or release notes, and does not require additional tooling for sharing or integration.
Weightscope is available as a CLI-first solution, installable via pip, and can be used for public model diffs and open-source exploration at no cost. The free tier includes access to the open-source CLI and library, as well as public diffs on HuggingFace Spaces for models up to 2 GB, and permits commercial use. Paid plans are available for users needing private checkpoint comparisons, hosted history, or organizational collaboration. The Pro plan offers private diffs, hosted history, shareable links, no model size limit, and email support, while the Team plan adds up to five seats, a shared workspace, SSO integration, audit logs, and Slack support. If a subscription lapses, users retain access to previously created artifacts but are limited to the free tier for new ones.
By making model diffs transparent and easily shareable, Weightscope serves researchers, developers, and organizations seeking to inspect, compare, and track changes in AI model weights.
In the AI & ML space, weightscope takes a focused approach. Understanding, comparing, and tracking changes in AI model weights for transparency and reproducibility. weightscope is an open-source project aimed at AI researchers and ML engineers. The project is open source (MIT). The product ships for the web and the command line.
weightscope first shipped in 2026. Among its 5 catalogued features are model weight diffing, sparse tensor analysis, and delta-as-patch export.
Latest indexed changes and source events
Other apps tracked under the same category.
