SafeView

A terminal application to view safetensors files. SafeView provides a clean, interactive terminal interface for exploring safetensors files and Hugging Face models.

Features

• Interactive terminal UI for browsing tensors
• Detailed tensor information including shape, data type, and size
• Statistical information about tensor values including min, max, mean, standard deviation, quantile analysis (1st, 10th, 25th, 50th/median, 75th, 90th, 99th percentiles), and sparsity percentage - loaded on demand when a tensor is selected
• Value distribution histogram visualization with toggle between linear and logarithmic scales
• Support for local safetensors files and Hugging Face model repositories with automatic downloading
• Real-time search and filtering by tensor name
• Clean and intuitive Textual-based interface with tabbed view (Details and Histogram)
• Optimized loading - only metadata is loaded initially, tensor statistics shown when a tensor is selected
• Progress indicators when loading tensor statistics
• Data preview snippets for quick inspection
• File-level information including total size, tensor count, and total parameter count

Installation

From PyPI:

pip install safe-view

Using pip (from local directory):

pip install .

Using uv:

uv pip install .

Development mode:

If you want to run in development mode, you can install in editable mode:

pip install -e .

or with uv:

uv pip install -e .

Usage

After installation, you can run the application directly from the command line:

safe-view /path/to/your/file.safetensors

Or for a Hugging Face model:

safe-view Qwen/Qwen3-0.6B

For help:

safe-view --help

Controls

• q: Quit the application
• h, j, k, l or arrow keys: Navigate between tensors
• g: Go to top of the tensor list
• G: Go to bottom of the tensor list
• Ctrl+f / Ctrl+b: Page up/down
• /: Enter search mode to filter tensors by name
• Escape: Exit search mode
• x or Enter: Load and display detailed statistics for the selected tensor
• Ctrl+l: Toggle histogram between linear and logarithmic scale
• Click on a tensor in the left panel to view its details and statistics on the right

Requirements

• Python 3.9+
• Dependencies listed in pyproject.toml