• XSS.stack #1 – первый литературный журнал от юзеров форума

AIoT - which hardware? ИИ Интернет вещей - какое оборудование?

Отслеживать
qubit_matrix

qubit_matrix

floppy-диск
Пользователь
Регистрация
21.01.2025
Сообщения
5
Реакции
0
Hello,

the last few weeks I was working on two IoT projects, one project was to build a C++ backend http(s) server on embedded hardware and the other project was to build an encrypted long range (LoRa 915 MHz) C++ half-duplex chat, also on embedded hardware. It was very enjoyable to build these things - even though at the beginning I had issues with implementing TLS encryption into the http server but I could solve the issue. On the other project I had issues with half-duplex mode but as soon as I knew the hardware was only capable of half-duplex mode, I could solve the issue as well.

Now I want to combine the https backend or/and the LoRa chat I wrote with an AI LLM that can run on my embedded hardware. I'm using Raspberry Pico and I can extend it with a microSD extension board (I already have and tested it, it works well).

Are there any hardware expansion modules I can use to run AI models on the Pico or would I need to buy a Raspberry Pi 5 Model with the AI Hat+?

This is what I have currently:
- Raspberry Pico 2W
- OLED Display
- microSD extension board
- LoRa extension board
- GPIO extender

I can possibly get this:
- https://www.raspberrypi.com/products/raspberry-pi-5/?variant=raspberry-pi-5-16gb
- https://www.raspberrypi.com/products/ai-hat/?variant=ai-hat-plus-26

But it seems very expensive. I also thought about getting Nvidia Jetson or Google Coral but this is even more expensive. Maybe something from espressif?

I will be using the C++ LiteRT library: https://ai.google.dev/edge/litert

Do you have a recommendation about hardware I can use? My budget is constrained, I wish I could spend 300$ but my limit currently is 150$.

------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Привет,


Последние несколько недель, когда я работал над двумя проектами IoT, один проект состоял в том, чтобы построить сервер HTTP (S) C ++ на встроенном аппаратном обеспечении, а другой проект состоял в том, чтобы создать зашифрованный на полуспутный чат C ++ C ++ (LORA 915 МГц), Также на встроенное оборудование. Было очень приятно создавать эти вещи - хотя в начале у меня были проблемы с внедрением шифрования TLS в HTTP -сервер, но я мог решить проблему. В другом проекте у меня были проблемы с полудуплексным режимом, но как только я узнал, что аппаратное обеспечение способно только к полудуплексному режиму, я также мог решить проблему.


Теперь я хочу объединить бэкэнд https или/и чат Lora, который я написал с помощью AI LLM, который может работать на моем встроенном оборудовании. Я использую Raspberry Pico, и я могу расширить его с помощью платы для расширения MicroSD (у меня уже есть и проверил, это работает хорошо).


Существуют ли какие -либо модули расширения аппаратного обеспечения, которые я могу использовать для запуска моделей ИИ на PICO, или мне нужно было купить модель Raspberry Pi 5 с AI HAT+?


Это то, что у меня есть в настоящее время:

- Raspberry Pico 2w

- OLED -дисплей

- Доска для расширения MicroSD

- LORA Extension Board

- gpio extender


Я могу получить это:

-https://www.raspberrypi.com/products/raspberry-pi-5/?variant=raspberry-pi-5-16gb

-https://www.raspberrypi.com/products/ai-hat/?variant=ai-hat-lus-26


Но это кажется очень дорогим. Я также думал о том, чтобы получить Nvidia Jetson или Google Coral, но это еще дороже. Может что -то из Espressif?


Я буду использовать библиотеку C ++ Litert: https://ai.google.dev/edge/litert


У вас есть рекомендация об оборудовании, которое я могу использовать? Мой бюджет ограничен, я хотел бы потратить 300 $, но мой лимит в настоящее время составляет 150 долларов.
 
I'm afraid "AI LLM" and "embedded" do not get on well together.
 
the AI model that can run on small hardware will be very dumb and/or very slow.
you will not be able to run anything on Pi Pico, and to save money I would suggest to try alternatives like an Orange Pi instead of the Raspberry Pi.
 
Dread Pirate Roberts сказал(а):
the AI model that can run on small hardware will be very dumb and/or very slow.
you will not be able to run anything on Pi Pico
It'll beat you at:
Tic_tac_toe.svg.png


Any who

Tiny Machine Learning: Progress and Futures

Tiny Machine Learning (TinyML) is a new frontier of machine learning. By squeezing deep learning models into billions of IoT devices and microcontrollers (MCUs), we expand the scope of AI applications and enable ubiquitous intelligence. However, TinyML is challenging due to hardware constraints...
arxiv.org

Edge Impulse - The Leading Edge AI Platform

Edge Impulse is the leading development platform for machine learning on edge devices.
edgeimpulse.com
 
shrekushka сказал(а):
It'll beat you at:
Посмотреть вложение 102676

Anyone

Tiny Machine Learning: Progress and Futures

Tiny Machine Learning (TinyML) is a new frontier of machine learning. By squeezing deep learning models into billions of IoT devices and microcontrollers (MCUs), we expand the scope of AI applications and enable ubiquitous intelligence. However, TinyML is challenging due to hardware constraints...
arxiv.org

Edge Impulse - The Leading Edge AI Platform

Edge Impulse is the leading development platform for machine learning on edge devices.
edgeimpulse.com
There's probably a research paper on the feasibility of intravenous seminal fluid as an alternative to regenerative stem cell treatment on Arxiv, but that doesn't make its application any less comical or practical =)
 
DimmuBurgor сказал(а):
There's probably a research paper on the feasibility of intravenous seminal fluid as an alternative to regenerative stem cell treatment on Arxiv,
Asking for a friend...
DimmuBurgor сказал(а):
but that doesn't make its application any less comical or practical =)
Everyone needs a hobby.

Now, with the arXiv reference, arXiv adds nothing of value to that paper. It's not a research paper. More of a summary/review of existing literature. It might as well been a WordPress blog.
Regardless, you bring up a good point about arXiv or any major of the pre-print servers: information warfare has been deployed on arXiv as well. The signal-to-noise ratio is ridiculously and dangerously low.

A little hyperbole here:
12.jpg
 
shrekushka сказал(а):
Asking for a friend...

Everyone needs a hobby.

Now, with the arXiv reference, arXiv adds nothing of value to that paper. It's not a research paper. More of a summary/review of existing literature. It might as well be a WordPress blog.
Regardless, you bring up a good point about arXiv or any major of the pre-print servers: information warfare has been deployed on arXiv as well. The signal-to-noise ratio is ridiculously and dangerously low.

A little hyperbole here:
Посмотреть вложение 102726
I'm going to need to see the reading on your oscilloscope, sir. The needle in the haystack is hanging out in vein, while my mind's bending spoons...

“Semenly” Harmless Back Pain: An Unusual Presentation of a Subcutaneous Abscess - PubMed

Aims We report, with review of the literature, the case of a patient who developed a subcutaneous abscess after intravenously injecting his own semen in an attempt to treat longstanding back pain. He had devised this “cure” independent of medical advice. Methods A review of EMBASE, PubMed...
pubmed.ncbi.nlm.nih.gov
 
Последнее редактирование:
why you want an "embedded" computer on ARM architecture instead of a "normal" x86 computer in a small form factor like Intel NUC or any "nettop"?
 
admin


Напишите ответ...
  • Вставить:
Прикрепить файлы
Верх