getting-started.md

Getting started

Prerequisites

TTNN Visualizer requires a preexisting report to analyze.

Follow instructions for TT-Metal and TT-NN

Sample config:

export TTNN_CONFIG_OVERRIDES='{
    "enable_fast_runtime_mode": false,
    "enable_logging": true,
    "report_name": "TODO ADD NAME",
    "enable_graph_report": true,
    "enable_detailed_buffer_report": true,
    "enable_detailed_tensor_report": false,
    "enable_comparison_mode": false
}'

To run a test with custom input data, you can use the following command:

pytest --disable-warnings --input-path="path/to/input.json" path/to/test_file.py::test_function[param]

For more information please refer to TT-Metalium, TT-NN and TT-NN models documentation.

The final output should be a folder including at least a config.json and a db.sqlite file.

Performance traces

TTNN Visualizer supports the reading of TT Metal performance traces. The expected output should be a folder container at least profile_log_device.csv and another csv with the performance results.

Consult the TT Metal documentation on how to generate a performance trace.

Installing as a Python Wheel

The application is designed to run on user local system and has python requirement of 3.12.3.

Download the wheel file from the releases page and install using pip install release_name.whl.

Running the application

After installation run ttnn-visualizer to start the application. If the app does not open automatically in your browser you can navigate to http://0.0.0.0:8000.

Docker

The application can also be alternatively installed and run via Docker.