What is ModelDiff?
ModelDiff is the world's first standalone browser-based tool for comparing AI model files. It lets you drag and drop two model files (.gguf or .safetensors) and instantly see their differences — tensors, metadata, quantization types, and size analysis.
Why ModelDiff?
The local LLM community frequently works with quantized models, fine-tunes, and merges. Before ModelDiff, there was no easy way to verify:
- Did my quantization change the model structure?
- What tensors changed in my fine-tune?
- Are these two models actually different?
- How much space does each quantization type take?
ModelDiff solves this. 100% client-side, no uploads, no server processing. Your files never leave your machine.
Supported Formats
GGUF (.gguf)
GGUF is the binary format used by llama.cpp and ollama. ModelDiff parses the header, metadata, and tensor info without loading the actual tensor data. This makes it fast even for 100GB+ files.
SafeTensors (.safetensors)
SafeTensors is a simple, safe format used by HuggingFace. It stores tensors with a JSON header. ModelDiff reads only the header, not the tensor data.
Cross-Format Comparison
You can compare GGUF vs SafeTensors. ModelDiff matches tensors by name and shows you the structural differences.
How It Works
- Drag & Drop: Drop two model files into the side-by-side zones.
- Parse: ModelDiff reads only the headers using File.slice() — no full file loading.
- Compare: It matches tensors by name, categorizes changes (added/removed/type changed/shape changed), and computes metadata diffs.
- Visualize: View results in 4 tabs: Summary, Metadata Diff, Tensor Diff, and Size Analysis with interactive charts.
Technology
Built with pure HTML/CSS/JavaScript. No frameworks, no build step, no dependencies except ECharts for visualizations. Works offline once loaded.
Privacy
Your files are processed entirely in your browser. Nothing is uploaded. See our Privacy Policy for details.
Contact
Questions, feedback, or bug reports? Email nullkit.dev@outlook.com.