Iskren-project-1 - Deployment

Iskren Zhechev / Iskren-project-1 Public

Target: Nordic Thingy:53 Clone this project

Configure your deployment

You can deploy your impulse to any device. This makes the model run without an internet connection, minimizes latency, and runs with minimal power consumption. Read more.

Run your model as a Docker container

To run your model as a container with an HTTP interface on NVIDIA Jetson's GPUs (JetPack 4.6.x), use:

Container:

public.ecr.aws/g7a8t7v6/inference-container-jetson:b7e0dc65374f7562ff72a40dd4c6623c8df0b5a9

Arguments:

--api-key ei_b955dc341461faddfebdc7f59293fe35abc36f70232715f46ca13f102eaa65e5 --run-http-server 1337

Ports to expose:

1337

For example, in a one-liner locally:

docker run --rm -it --runtime=nvidia --gpus all \
    -p 1337:1337 \
    public.ecr.aws/g7a8t7v6/inference-container-jetson:b7e0dc65374f7562ff72a40dd4c6623c8df0b5a9 \
        --api-key ei_b955dc341461faddfebdc7f59293fe35abc36f70232715f46ca13f102eaa65e5 \
        --run-http-server 1337

This automatically builds and downloads the latest model (incl. hardware optimizations), and runs an HTTP endpoint at http://localhost:1337 with instructions.

Read the docs for information, like bundling in your model inside the container and selecting extra hardware optimizations.

Run your model as a Docker container

To run your model as a container with an HTTP interface on NVIDIA Jetson Orin's GPUs (JetPack 5.1.x), use:

Container:

public.ecr.aws/g7a8t7v6/inference-container-jetson-orin:fc4ef2bb6ce3e51bd1a391551d0d57ca745d8257

Arguments:

--api-key ei_b955dc341461faddfebdc7f59293fe35abc36f70232715f46ca13f102eaa65e5 --run-http-server 1337

Ports to expose:

1337

For example, in a one-liner locally:

docker run --rm -it --runtime=nvidia --gpus all \
    -p 1337:1337 \
    public.ecr.aws/g7a8t7v6/inference-container-jetson-orin:fc4ef2bb6ce3e51bd1a391551d0d57ca745d8257 \
        --api-key ei_b955dc341461faddfebdc7f59293fe35abc36f70232715f46ca13f102eaa65e5 \
        --run-http-server 1337

This automatically builds and downloads the latest model (incl. hardware optimizations), and runs an HTTP endpoint at http://localhost:1337 with instructions.

Read the docs for information, like bundling in your model inside the container and selecting extra hardware optimizations.

Run your model as a Docker container

To run your model as a container with an HTTP interface on NVIDIA Jetson Orin's GPUs (JetPack 6.0), use:

Container:

public.ecr.aws/g7a8t7v6/inference-container-jetson-orin-6-0:bebbe4e4fd1d6ced9b04da167c86411c50d2dc48

Arguments:

--api-key ei_b955dc341461faddfebdc7f59293fe35abc36f70232715f46ca13f102eaa65e5 --run-http-server 1337

Ports to expose:

1337

For example, in a one-liner locally:

docker run --rm -it --runtime=nvidia --gpus all \
    -p 1337:1337 \
    public.ecr.aws/g7a8t7v6/inference-container-jetson-orin-6-0:bebbe4e4fd1d6ced9b04da167c86411c50d2dc48 \
        --api-key ei_b955dc341461faddfebdc7f59293fe35abc36f70232715f46ca13f102eaa65e5 \
        --run-http-server 1337

This automatically builds and downloads the latest model (incl. hardware optimizations), and runs an HTTP endpoint at http://localhost:1337 with instructions.

Read the docs for information, like bundling in your model inside the container and selecting extra hardware optimizations.

Model Optimizations

Model optimizations can increase on-device performance but may reduce accuracy.

Quantized (int8)


Latency
Ram
Flash
Accuracy

MFE	Classifier	Total
108 ms.	4 ms.	112 ms.
19.8K	10.6K	19.8K
-	33.1K	-
		-

Unoptimized (float32)


Latency
Ram
Flash
Accuracy

MFE	Classifier	Total
108 ms.	152 ms.	260 ms.
19.8K	32.9K	32.9K
-	32.9K	-
		-

To compare model accuracy, clone this project and run model testing.

Estimate for Nordic Thingy:53.

Clone this project to deploy this impulse.

Configure your deployment

Deploy to any Linux-based development board

Run your model as a Docker container

Run your model as a Docker container

Run your model as a Docker container

Run your model as a Docker container

Model Optimizations

Quantized (int8)

Unoptimized (float32)

Latest build

Run this model

Build output

Model testing output

Configure your deployment

Default Deployment

C++ library

Deploy to any Linux-based development board

Run your model as a Docker container

Run your model as a Docker container

Run your model as a Docker container

Run your model as a Docker container

Model Optimizations

EON™ Compiler

EON™ Compiler

TensorFlow Lite

Quantized (int8)

Unoptimized (float32)

Latest build

Run this model

Build output

Model testing output