Quick Start

TT-NN Visualizer can be installed from PyPI:

pip install ttnn-visualizer
# or
pipx install ttnn-visualizer
# or
uv tool install ttnn-visualizer

After installation run ttnn-visualizer to start the application.

It is recommended to do this within an isolated/virtual environment. The minimum Python version is 3.10.

Please see the install guide guide for further information on getting up and running with TT-NN Visualizer.

If you want to test out TT-NN Visualizer you can try some of the sample data. See loading data for instructions on how to use this.

Features

For the latest updates and features, please see releases.

Reports

• Upload reports from the local file system or sync remotely via SSH
• Switch seamlessly between previously uploaded or synced reports
• Run multiple instances of the application concurrently with different data
• Set data ranges for both memory and performance traces
• Display physical topology and configuration of Tenstorrent chip clusters

Operations

• Filterable list of all operations in the model
• Interactive memory and tensor visualizations, including per core allocations, memory layout, allocation over time
• Input/output tensors details per operation including allocation details per core
• Navigable device operation tree with associated buffers and circular buffers

Tensors

• List of tensor details filterable by buffer type
• Flagging of high consumer or late deallocated tensors

Buffers

• Visual overview of all buffers for the entire model run by L1 or DRAM memory
• Toggle additional overlays such as memory layouts or late deallocated tensors
• Ease of navigation to the relevant operation
• Track a specific buffer in the data across the application
• Filterable table view for a more schematic look at buffers

Graph

• Interactive model graph view showing all operations and connecting tensors
• Filter out deallocated operations
• Find all operations by name

Performance

• Integration with tt-perf-report and rendering of performance analysis
• Interactive charts and tables
• Multiple filtering options of performance data
• Compare multiple performance traces

NPE

• Network-on-chip performance estimator (NPE) for Tenstorrent Tensix-based devices
• Dedicated NPE visualizations: zones, transfers, congestion, timelines with elaborate filtering capability

Demo

Application demo

https://github.com/user-attachments/assets/4e51a636-c6d6-46df-bf34-a06bca13c0b3

Sample reports

You may test the application using the following sample reports.

Unzip the files into their own directories and select them with the local folder selector, or load the NPE data on the /npe route.

Segformer encoder memory report

Segformer decoder memory report

Llama mlp memory + performance report

N300 llama memory + performance report with NPE data + cluster description

NPE report

T3K synthetic synthetic_t3k_small.json.zip

Contributing

How to run TT-NN Visualizer from source.