NPY and NPZ Viewer Online
Drop a NumPy .npy or .npz file to see the shape and dtype of every array, its stats, its first rows and a simple chart. No Python, no account and no upload.
NPZ Viewer & NPY Viewer for VS Code and Cursor, free
- Opens .npy and .npz files, PyTorch .pt, .pth and .ckpt checkpoints, and GGUF and Safetensors models, with no Python.
- Stats with median and percentiles, histograms, heatmaps, arrays drawn as pictures, and a values table that pages through any tensor.
- Saves a tensor or a slice as .npy or CSV, and reloads when your training loop writes the file again.
Optional Zovo Pro, USD 19 once: compare two checkpoints, and scan all tensors for NaN, Inf and constant values. 3 free tries for each.
Drop a .npy or .npz file here
or
NumPy .npy and .npz (np.save, np.savez and np.savez_compressed), any size. Big arrays are read from the start.
| Array | Type | Shape | Size |
|---|
First rows
Open this in VS Code or Cursor
The free NumPy Viewer extension shows your arrays inside VS Code and Cursor with a click on the file. It also opens PyTorch .pt and .pth checkpoints and GGUF and Safetensors models, draws arrays as pictures, pages through every value and saves a tensor as .npy or CSV.
Free, MIT license. About NumPy Viewer for VS Code.
Zovo Pro
Compare two checkpoints and scan all tensors. Zovo Pro, USD 19 once.
In the NumPy Viewer extension, Zovo Pro compares two checkpoints tensor by tensor (max and mean difference, RMSE, cosine similarity, shape changes) and reads every value of every tensor to flag NaN, Inf, all-zero and constant tensors. Both reports save as CSV, Markdown or JSON. Each Pro tool works 3 times for free. One payment covers the Pro tools in 22 Zovo extensions. Everything that is free stays free.
After a card payment you get a key on the page. With GitHub Sponsors you sign in from the extension. Details on the Zovo Pro page.
Related tools
- NPZ & NPY Viewer, PyTorch & Safetensors on the VS Code Marketplace (also on Open VSX). Shows arrays, shapes and values from .npy, .npz, .pth, GGUF and safetensors files.
- NPZ, NPY, PyTorch and GGUF Viewer. Shows tensors, shapes, stats and heatmaps from model files in VS Code.
- Pickle Viewer & PKL Viewer. Reads .pkl files as data in VS Code, with no Python.
- Parquet Viewer & Parquet Explorer. Opens large Parquet files with nested data in VS Code.
- Parquet Viewer Online, free in your browser.
- CSV Viewer Online, free in your browser.
- Hex Viewer Online, free in your browser.
Questions
Is my file uploaded anywhere?
No. The page reads the file inside this tab with the File API. Its Content Security Policy sets connect-src to none, so the page cannot send your arrays anywhere. One small library (fflate, to unpack compressed .npz files) loads from cdn.jsdelivr.net at a pinned version, and your browser checks its hash before running it. Your file is never sent there.
Which files does it open?
Files written by np.save (.npy format 1.0, 2.0 and 3.0) and by np.savez and np.savez_compressed (.npz, stored or deflated, including ZIP64). Types: bool, int and uint of 8 to 64 bits, float16, float32 and float64, complex64 and complex128, datetime64 and timedelta64, unicode and byte strings, little or big endian, C or Fortran order.
How big an array can it show?
The page has no file size limit of its own. It reads the headers first, then up to 64 MB of the array you click. For a bigger array, the stats, the chart and the first rows cover its first 64 MB, and the page says so. The free NumPy Viewer for VS Code reads every value.
What do the charts show?
The line chart shows the values in file order; when there are more values than pixels, the line is the mean of each pixel column and a light band shows its min and max. The histogram counts values in 40 equal bins. For 2-D and bigger arrays the heatmap shows the last two axes, and a slice box steps through the leading axes, so a stack of images shows one image at a time. Values below and above 0 get a blue and red scale with gray at 0.
Is a .npy file with Python objects safe to open?
Yes. Arrays of Python objects (dtype O) are stored as a pickle, which can run code when numpy loads it with allow_pickle=True. The page never runs a pickle: it lists such an array and says why it shows no values. The header is read as text and never run, and every value is written as plain text, never as HTML.
Does it work offline?
Yes. After the first visit the page keeps a copy of its own files and of the pinned library, so it opens arrays with no connection.
What is free and what is Zovo Pro?
Everything on this page is free. In the NumPy Viewer extension for VS Code, every format, stats, histograms, heatmaps, pictures, the values table and Save as .npy or CSV are free too. Comparing two checkpoints and scanning all tensors are Zovo Pro, with 3 free tries each. One USD 19 payment covers the Pro tools in every Zovo extension.