diffai
"Binary files differ" is not enough
Shows meaningful diffs for ML model files. Learning rate changes, gradient health, layer additions and deletions at a glance.
"Model changes visible at a glance"
— ML Engineer
"Convenient for tracking training progress"
— Researcher
Features
11 Auto Analyses
Learning rate, optimizer, loss, accuracy, model version, gradient, quantization, convergence, activation, attention, ensemble.
Tensor Statistics
Compare mean, std, min, max. Detects shape changes and parameter additions/deletions.
4 Formats
PyTorch (.pt/.pth), Safetensors, NumPy (.npy/.npz), MATLAB (.mat).
JSON/YAML Output
--output json/yaml for machine-readable output. Detect model changes in CI/CD.