Using Probe
What is NetsPresso Probe?
NetsPresso Probe is a browser-based graph visualizer designed for AI engineers who need to quickly inspect, compare, and understand neural network models.
It provides fast loading performance even for large LLM models and offers a powerful Diff feature for structural comparison before and after optimization.
If you’d like to try Optimization Studio, please visit the link below:
NetsPresso Probe
Target Users
- AI engineers and researchers performing model architecture analysis
- Users who need to compare pre- and post-optimized models
Key Features
1. Diff Mode
Compare two models side-by-side to visually identify structural differences such as added, removed, or modified nodes.
- Topology Visualization: Verify optimization patterns by visualizing topology changes before and after graph optimization.
- Multiple View Layouts: Supports various comparison layouts including Basic, Parallel, Derived, and Aligned views to suit different analysis needs.
- Summary Indicators: Provides quick insights with indicators for similarity percentage, total node count, and added/deleted nodes.
2. Node Coloring
Assign custom colors to operators for faster structural understanding and analysis.
- Highlight Important Ops: Highlight operations that are important to your experiment by assigning custom colors or using predefined sets.
- Conditional Attribute Coloring: Apply custom colors by specifying conditions for desired attribute or metadata values, allowing for precise visual filtering.
3. Synced Control
Enabling Synced Control synchronizes the viewpoint between the source and target models, making the exploration of structural differences and localized topology changes significantly more efficient.
4. Fast Loading Performance for LLMs & Large Graphs
Netspresso Probe is engineered with an optimized parsing engine designed to handle massive computation graphs. While standard visualization tools often suffer from lagging or crashes when loading LLMs, Netspresso Probe ensures smooth and stable rendering.
Performance Benchmark
The table below compares the loading performance of Netspresso Probe against a standard model viewer (Netron). Netspresso Probe demonstrates superior speed and stability, particularly with complex architectures like Gemma-3 and GPT-2.
| Base Model | Model Name | NetsPresso Probe | Other Model Viewers | Performance Gain |
|---|---|---|---|---|
| Gemma 3 | gemma3n-E2B-it | 14.5s | Crashed | Successful Rendering |
| GPT-2 | openai-community/gpt2-large | 6.5s | 11.4s | ~1.75x faster |
Updated 3 days ago
