Don't be afraid to show us a bit of leg from time to time.
It takes all kinds.
Don't be afraid to show us a bit of leg from time to time.
It takes all kinds.
Remember that, like sgt. Schultz, I know nothing about Transformers.Just having a little look around and the orange xxxx’s mark the spot where in the past week and interesting update was made to AKIDA models. @Diogenese can tell us if this amendment has any significance and almost forgot since when has AKD500 been available. Peter van der Made said this might be used in white goods:
Upgrade to akida/cnn2snn 2.2.6 and akida_models 1.1.8
last week
ktsiknos-brainchip![]()
2.2.6-doc-1
d334eea
Upgrade to akida/cnn2snn 2.2.6 and akida_models 1.1.8
Latest
Update akida and cnn2snn to version 2.2.6
New features
- [akida] Upgrade to quantizeml 0.0.13
- [akida] Attention layer
- [akida] Identify AKD500 devices
- [engine] Move mesh scan to host library
API changes
- [engine] toggle_learn must be called instead of program(p,learn_enabled)
- [engine] set_batch_size allows to preallocate inputs
Bug fixes
- [engine] Memory can grow indefinitely if queueing is faster than processing
Update akida_models to 1.1.8
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
- updated CNN2SNN minimal required version to 2.2.6 and QuantizeML to 0.0.13
- VWW model and training pipeline refactored and aligned with TinyML
- Layer names in almost all models have been updated in preparation for quantization with QuantizeML
- Tabular data models and tools have been removed from the package
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
- Transformers pretrained models updated to 4-bits
- Introduced calibration utils in training toolset
- KWS and ImageNet training scripts now offer a "calibrate" CLI action
- ImageNet training script will now automatically restore the best weights after training
Documentation update
- dropped quantizeml API details for now
![]()
Releases · Brainchip-Inc/akida_examples
Brainchip Akida Neuromorphic System-on-Chip examples and documentation. - Brainchip-Inc/akida_examplesgithub.com
AI Backing AI. ClassicI proposed this question on the openai CHATbot.. very interesting to read the response..
“How will Brainchip akida revolutionise the AI sector”
Brainchip Akida is a neuromorphic system-on-chip (SoC) that has been designed to enable artificial intelligence (AI) and machine learning (ML) capabilities in edge devices. The Akida SoC is based on a novel spiking neural network (SNN) architecture that is inspired by the way the brain works. According to Brainchip, the Akida SoC can perform a wide range of AI and ML tasks, including image recognition, object detection, and natural language processing, with high accuracy and low power consumption.
The potential benefits of the Akida SoC for the AI sector are significant. By enabling AI and ML capabilities on edge devices, the Akida SoC could potentially revolutionize the way that AI is deployed and used. For example, the Akida SoC could enable the development of AI-powered devices that can operate independently, without the need for a connection to a central server or cloud computing infrastructure. This could make AI more accessible and widely available, and enable a wide range of new applications and use cases.
It is important to note that the Akida SoC is still a relatively new technology, and it remains to be seen how it will be adopted and used in the AI sector. However, the potential benefits of the Akida SoC are significant, and it is likely that it will play a significant role in the future of AI and ML.
Legend, thanks Tech.Good morning from another beautiful day in Perth...30c already at 8am
I'm not too sure that, the above quote is 100% accurate, other companies' chips with our IP embedded would be a lot more
accurate.
We simply don't supply chips, IP in blocks is how I understand it to be moving forward, I also understand what she is implying and maybe I'm being a little pedantic.
And for Santa's little helpers still shaking our Christmas Tree, the only thing falling off is fluff, which we don't deal in anymore.
Our tree will never fall over no matter how much shaking you give it, why, because our foundations are rock solid.
See you on the other side of CES 2023, I believe that my neighbour has a meeting arranged with the Brainchip team in Las Vegas to discuss the possibly of having her engineers in the South African mining industry work in with our guys to develop Akida technology for underground mining in the areas of gas sensing, predictive maintenance, vibration analysis etc.
I'll ask her to take some photos if possible while at CES.
Tech x![]()
OKAY thanks for being there yet again and yes at least on my reading Transformers go hand in hand with LSTM.Remember that, like sgt. Schultz, I know nothing about Transformers.
As I understand it:
Update akida_models to 1.1.8
- Transformers pretrained models updated to 4-bits
means the model libraries (speech was the initial application of Transformers) have been updated.
There are about 44 phonemes in English, but dialect and accent will multiply this.
The updating was to convert the libraries to 4-bit bytes (from ... 8-bits?) so they are compatible with Akida NPUs.
This is only changing some numbers in memory - no silicon was harmed in performing this update.
Remembering that we are selling Akida IP, implementing Transformers in Akida may be as simple as updating the model library and designing the appropriate NPU configuration, the configuration being implemented by the ARM Cortex M MPU as part of the set up procedure.
Now I suppose you do need LSTM for transformers, so this will require some additional memory for previous words/sentences, so there may be some collateral damage to the silicon at this stage.
So I wonder if we have had to go back to the drawingboard after implementing LSTM to update the silicon for Transformers which are a very recent phenomenon for Akida.
We could sign a deal with the biggest company ever and the shorters would still short the crap out of it, the names mentioned this share shouldn't be at this price
So if Intel need us let's hope where going to charge them the MaxIts a very delicate situation.
Imagine your supervisors and bosses Investers spent billions to develop a less superior product!
You don't just go oh we were wrong and let's go this way. It takes a plan and time to steer that giant vessel a different direction. You know how many jobs and manager roles and friends would have been booted with a sharp reaction. It's a big political game also.
Know Intel will make money and be in the chip race for edge AI. It's only time that the rest join.
I know we would love to hear them sing AKIDA but they won't not yet.
This partnership was the last DD I needed to really feel confident in BRN succeeding. As Arm, Merc, Renesas, Mega Chips and more weren't
This is a sign how quick others will adopt.
IMO
Merry Christmas Happy Holidays folks.
Great post FF, much appreciated by everyone here.One reason I can think of is if Mike Davies/Intel says this loud and clear it is a signal to every other company that they don’t need Intel and can go straight to Brainchip.
At the moment they can continue to try and control the narrative that Loihi is the way to neuromorphic ascendancy and Brainchip AKIDA IP is a niche chip for limited use cases and if you really want it they can supply it.
But Brainchip is building itself into as many ecosystems as it possibly can in much the same way as a virus moves through a community.
Renesas is tapping out at the low end while at the same time Edge Impulse is comparing AKIDA with GPUs.
Peter van der Made has stated that the market has no understanding of the significance of MegaChips to the future success of Brainchip.
Prophesee and the use cases it referenced for an AKIDA Prophesee event based intelligent sensor picks up many of the industry directions referenced in the report I posted above.
At the same time ARM the chip supply monster is promoting AKIDA across virtually every industrial use case.
NASA is clearly exploring the use of AKIDA as an essential element of deep space exploration and DARPA is deeply imbedding the AKIDA technology in radar and other use cases via ISL and others.
Then Biotome is one of the known medical research companies exploring the use of AKIDA for this industry.
Mercedes Benz extolling the AKIDA advantage over all current technology options in the automotive industry.
Remember Alex the Rocket Scientist stating that he refers to autonomous vehicles as robots because that is what they are technically. Mercedes Benz has not simply extolled the benefits of using AKIDA for cars needing voice recognition but for every single robotic use case from drones to personal assistants.
Carnegie Mellon University and others are now teaching AKIDA Science Fiction to the next generation of technology entrepreneurs and innovators who will populated the research labs and offices of the technology giants.
Brainchip’s Board and Management are brilliantly in my opinion creating an environment where AKIDA is ubiquitous and if you are not involved in some way then you are not on the right technology page.
In my opinion Intel had no choice but to join and will try to control an uncontrollable narrative which has Renesas offering AKIDA for making low end MCU’s smart and Edge Impulse describing AKIDA as a threat to GPUs and the stuff of Science Fiction.
The eventual release of AKIDA next gen into the established and growing ecosystem will be like hitting the nitrous switch on a dragster.
My opinion only DYOR
FF
AKIDA BALLISTA
Love any dot joining and happy to analyse and review new revelations..but..a link to a Japanese dog breed? Farfetched, methinks. Isaac's sacrifice in Hebrew is 'Akedat Yitzchak'..as he was nearly 'spiked' in sacrifice.Surely not? View attachment 25199
View attachment 25223
Keep pulling everyone. We’re nearly there.
Can’t wait till the earth gives way and we get our giant carrot!
Merry Christmas to all x
Hi @Fact FinderYes the timetable was always for the IP to be released second half 2022 and if you have a look at Sean Hehir's AGM presentation he states it is going to happen and that a reference chip is likely to be produced next year.
I do not want to prejudice your interpretation of his exact words but they left me wondering if the IP release was more flexible than before 31.12.22 as I had thought from statements by Peter van der Made, Anil Mankar and Ken Scarince it was a definite for 2022 until then.
Certainly they have purchased and paid for all the third party IP they required as this was reported in the half yearly report and clarified by Ken Scarince that the IP purchase covered the AKIDA 2.0 third party IP.
I know you know how to access the AGM presentation so I would be interested to hear your take on how to understand Sean Hehir on the release date for the IP.
If it is not released before 31.12.22 I would suspect it is a tactical decision related to a customer engagement as the IP was ready end of last year and it is the engineering that has been taking place this year. Going by AKD1000 the engineering clearly was not finalised when the IP was released to select customers back in 2019 and when it was made more generally available they were still working on the FPGA to ensure they could get the production of AKD1000 right the first time. Which thanks to Anil Mankar's brilliance in this area they did.
My opinion only DYOR
FF
AKIDA BALLISTA
Wish you all a Merry Christmas and a good start into the new year! May we see next year a better market situation than this year! Stay healthy and happy![]()
Best
7