beetlebox

Пікірлер

@juanpauloazaelpalacios-vil5422 Ай бұрын

Hi, what software are you using to run Vivado on macOS? Do you have an M-series MacBook?

@beetlebox4858 Ай бұрын

Vivado runs on our Ubuntu server that we remote connect to.

@embetrix 26 күн бұрын

What advantages does you CI Workflow Bring vs Classical GitLab/Bitbucket/Jenkins one ?

@beetlebox4858 26 күн бұрын

@@embetrix BeetleboxCI workflows are able to uniquely manage and orchestrate external devices. The software makes it simple to reprogram, connect to and run tests on-device. The workflow even manages the availability of a device to ensure two different tests do not clash with one another. This particular tutorial doesn't highlight these features so much because there isn't really an external device. However, we made it to provide the best education and support of any CI service in regards to embedded systems and hardware development.

@juanpauloazaelpalacios-vil5422 Ай бұрын

currently using Brevitas/FINN frameworks to deploy a tiny-yolo model on a Kria board! Love the hardware design challenge of it all

@beetlebox4858 Ай бұрын

Sounds very cool. We are looking at developing a Vitis AI and MLOps video as well.

@juanpauloazaelpalacios-vil5422 Ай бұрын

@@beetlebox4858 Looking forward to it, thanks for putting in the work in these videos by the way! High quality stuff!

@novantha1 Ай бұрын

An interesting expansion on custom AI models: There are lots of quantization schemes coming up that can't be employed to full efficiency on classical GPUs or CPUs, such as Bitnet 1.58, which is dependent on extremely efficient turnary operations which in contrast can be accelerated quite aggressively on FPGAs and ASICs.

@beetlebox4858 Ай бұрын

In the case of binary and ternary, FPGAs has huge advantage over GPUs because of their ability to exploit XOR operations. Problem is that accuracy drops badly below the 4 bit mark.

@novantha1 Ай бұрын

@@beetlebox4858 In the case Bitnet specifically the accuracy actually isn't terribly far off of full precision when trained with quantization aware training as described in the paper; the main issue is actually having to train the model from scratch, but it does work pretty well. And, as you noted, it's very efficient on FGPAs.

@souravjamwal77 Ай бұрын

I am always interested in applying current trends to the core or very base hardware. Like last year I built a custom lambda image that had low latency than an 8Gb EC2 instance with GPU. Thanks for the video.

@beetlebox4858 Ай бұрын

EC2 actually has some FPGA images called F1s that you can experiment with.

@souravjamwal77 Ай бұрын

@@beetlebox4858 thanks

@elab_DIY 3 ай бұрын

thaks u !!!

@dekev7503 3 ай бұрын

This Gem 💎 is very difficult to find. Just subbed to your channel, look forward to seeing more from you.

@beetlebox4858 3 ай бұрын

Thank you so much. More regular content has already been scheduled.

@dekev7503 3 ай бұрын

@@beetlebox4858 can’t wait

@JonasSolar 3 ай бұрын

How would hardware or sensor input work? Obviously nobody running around pushing buttons ;)

@beetlebox4858 3 ай бұрын

It depends on the specific hardware or sensor. Sometimes it is just as simple as connecting the microcontroller and sensor to the same breadboard. You can see an example in our STM32 demo where we use a I2C LED screen. All we do is just cycle the power and check the driver is responding properly before we write to the screen. In other cases, we have created more sophisticated setups where we use a seperate microcontroller to control a test kit with all the various sensors and hardware attached. The test kit and the device can then be synced together through CI. We do this setup if there are very specific test requirements like needing to know exactly how the system will respond if a sensor goes down.