Inspect and visualize local .npy, .npz, .pt, and .pth tensors directly in your browser.
Slice, tile, reshape, compare, and hover-inspect values with local stats and keyboard control.
| Name | Shape | Channels | Slice | Dtype | Min/Max | Mean ± Std | Counts | Actions |
|---|
Load NumPy and PyTorch tensor files and review shapes, dtypes, storage order, and metadata at a glance.
Visualize tensors to confirm patterns, and hover to inspect exact values and indices.
Hover stats include min/max, mean ± std, and counts of zeros, NaNs, and ±Infs.
Choose x/y axes, tile extra dimensions, and flip channels-first/last layouts when auto-detection is wrong.
Configure which dimensions are sliced by index versus tiled, and reshape into new tensors.
NaN and Inf values render in high-contrast colors so anomalies stand out.
Use Page Up/Down to step through tensors, or Shift+Page Up/Down to jump to the next compatible view size for a visual diff.
Keep the same layout to compare tiles, slices, and channel order side by side.
Open NPZ or PT bundles, select just the tensors you need, and export them as individual NPY files.
Apply reshapes before export, without materializing full tensors in memory.
Pan and zoom in tile space to inspect fine structure, or jump to fit and 1:1 zoom.
Arrow keys select slice dimensions (up/down) and adjust slice indices (left/right).
Tensor Spy reads .npy, .npz, .pt, and .pth tensor files, including multi-tensor archives, non-image data, and major image formats.
Tensor data is processed in-browser and cached locally for fast revisits; nothing is uploaded or materialized in full.
Hover the canvas to read exact values and indices, and review per-slice and full-tensor stats for min/max, mean ± std, and counts of zeros, NaNs, and ±Infs.
Arrow keys change slice dimensions (up/down) and step indices (left/right), while Page Up/Down switches tensors and Shift+Page Up/Down jumps to compatible views.