Model Demos

This page demonstrates how to run the tt-metalium model demos using the tt-installer tool. The tt-installer can be used to download a container for running the tt-metalium demos. This container possesses a full build of the tt-metalium project, including the demo source code.

⚠️ NOTE: this page assumes you have already used tt-installer to install the system dependencies as shown in the starting guide

1. Downloading and installing the tt-metalium models container

Execute the following command to download and install the tt-metalium models container:

/bin/bash -c "$(curl -fsSL https://github.com/tenstorrent/tt-installer/releases/latest/download/install.sh)" --no-install-kmd --no-install-hugepages --no-install-metalium-container --install-metalium-models-container --no-install-tt-flash --no-install-tt-topology --update-firmware="off" --reboot-option="never" --mode-non-interactive

2. Starting the container and running a simple program

To use the models container, execute this command to create an interactive shell with all configuration taken care of:

tt-metalium-models

NOTE: this container is ephemeral so all changes made inside will be lost when the container is stopped

Run the following simple program to perform a simple exponentiation + matrix multiplication operation:

# this should be run inside the container
python ttnn/ttnn/examples/usage/run_op_on_device.py

3. Run more model demos

The full list of model demos to run can be found in the tt-metal GitHub repository. Scroll down that page to find the table of model demos for the following model types:

LLMs
Speech-To-Text
Diffusion
Image Classification
Vision Transformers
Object Detection
Segmentation
NLPs

Clicking on each model’s name will navigate to the demo page which contains all instructions for running that demo.

4. Exiting the container

To exit the container, simply run this command:

exit