safetensors Explorer

An interactive terminal-based explorer for safetensors and GGUF files, designed to help you visualize and navigate the structure of machine learning models.

Features

• 🔍 Interactive browsing of safetensors and GGUF file structures
• 📁 Hierarchical tree view with expandable/collapsible groups
• 🔎 Fuzzy search - instantly filter tensors with fuzzy matching using / key
• 🔢 Smart numeric sorting for layer numbers (e.g., layer.0, layer.1, layer.2, ..., layer.10)
• 📊 Tensor details including shape, data type, and size
• 🔗 Multi-file support - automatically merges multiple files into a unified view
• 📂 Directory support - explore entire model directories with automatic safetensors index detection
• 🌟 Glob pattern support - use wildcards to select multiple files (e.g., *.safetensors, model-*.gguf)
• 📏 Human-readable sizes (B, KB, MB, GB)
• ⌨️ Keyboard navigation for smooth exploration
• 🧠 GGUF support - view GGML format tensors with quantization types

Installation

From crates.io

cargo install safetensors_explorer

Prerequisites

• Rust (1.70 or later)

Build from source

git clone https://github.com/EricLBuehler/safetensors_explorer
cd safetensors_explorer
cargo build --release

Usage

Basic usage

# Explore a single safetensors file
safetensors_explorer model.safetensors

# Explore a GGUF file
safetensors_explorer model.gguf

# Or if building from source
cargo run -- model.safetensors
cargo run -- model.gguf

Directory exploration

# Explore all safetensors and GGUF files in a directory
safetensors_explorer /path/to/model/directory

# Recursively search subdirectories
safetensors_explorer -r /path/to/models

# The tool automatically detects and uses model.safetensors.index.json if present
safetensors_explorer /path/to/huggingface/model

Multi-file exploration

# Explore multiple files as a unified model
safetensors_explorer model-00001-of-00003.safetensors model-00002-of-00003.safetensors model-00003-of-00003.safetensors

# Mix safetensors and GGUF files
safetensors_explorer model.safetensors model.gguf

# Mix files and directories
safetensors_explorer model.safetensors /path/to/additional/models

Glob pattern support

# Use wildcards to select multiple files
safetensors_explorer *.safetensors

# Match files with specific patterns
safetensors_explorer model-*.gguf

# Match numbered checkpoint files
safetensors_explorer checkpoint-[0-9]*.safetensors

# Combine multiple patterns
safetensors_explorer *.safetensors *.gguf

# Mix glob patterns with explicit paths
safetensors_explorer model.safetensors checkpoint-*.safetensors

Keyboard Controls

Key	Action
↑ / ↓	Navigate up/down through the tree
Enter / Space	Expand/collapse groups, view tensor details
/	Enter search mode to filter tensors
Esc	Exit search mode
q	Quit the application (or exit search mode if active)
Ctrl+C	Force quit

Search Feature

Press / to enter search mode and start typing to filter tensors by name. The search:

• Uses fuzzy matching - find tensors even with typos or partial matches (e.g., "attnproj" will match "attn.c_proj.weight")
• Searches all tensors - not just visible ones, regardless of collapsed groups
• Shows results in a flat list with full tensor names
• Sorts by relevance - best matches appear first

Press Enter or Esc to exit search mode and return to the full tree view.

Example Output

SafeTensors Explorer - model.safetensors (1/1)
Use ↑/↓ to navigate, Enter/Space to expand/collapse, q to quit
================================================================================

▼ 📁 transformer (123 tensors, 1.2 GB)
  ▼ 📁 h (120 tensors, 1.1 GB)
    ▼ 📁 0 (5 tensors, 45.2 MB)
      📄 attn.c_attn.weight [Float16, (4096, 3072), 25.2 MB]
      📄 attn.c_proj.weight [Float16, (1024, 4096), 8.4 MB]
      📄 ln_1.weight [Float16, (4096,), 8.2 KB]
      📄 mlp.c_fc.weight [Float16, (4096, 11008), 90.1 MB]
      📄 mlp.c_proj.weight [Float16, (11008, 4096), 90.1 MB]
    ▶ 📁 1 (5 tensors, 45.2 MB)
    ▶ 📁 2 (5 tensors, 45.2 MB)
    ...
    ▶ 📁 31 (5 tensors, 45.2 MB)
  📄 ln_f.weight [Float16, (4096,), 8.2 KB]
  📄 wte.weight [Float16, (151936, 4096), 1.2 GB]

Selected: 1/342 | Scroll: 0

How It Works

1. Path Resolution: Automatically discovers safetensors files from files, directories, or safetensors index files
2. File Loading: Loads one or more safetensors files and extracts tensor metadata
3. Tree Building: Organizes tensors into a hierarchical structure based on their names (split by '.')
4. Smart Sorting: Uses natural sorting to handle numeric components correctly
5. Interactive Display: Renders the tree with expansion/collapse functionality
6. Tensor Details: Shows detailed information when selecting individual tensors

Technical Details

Supported Formats

• safetensors files (.safetensors)
• GGUF files (.gguf) with GGML tensor types including quantized formats
• safetensors index files (model.safetensors.index.json)
• Directory scanning with recursive search option
• All tensor data types supported by the safetensors and GGML formats

Performance

• Memory efficient: Only loads tensor metadata, not the actual tensor data
• Fast startup: Optimized for quick exploration of large models
• Responsive UI: Smooth navigation even with thousands of tensors

Dependencies

• safetensors - For reading safetensors files
• gguf - For reading GGUF files
• crossterm - For terminal UI and keyboard input
• clap - For command-line argument parsing
• anyhow - For error handling
• serde_json - For parsing safetensors index files
• glob - For directory pattern matching

Contributing

Contributions are welcome! Please feel free to submit issues or pull requests.