elephantum

I see, in order to benefit, model has to be quantized. It is not super clear which kinds of quantization are supported. Both Fp16 and Int8?

Do the same optimizations apply to tensorflow/tensorflow serving?

Is this a drop in solution that works with every existing tflite model?

I killed at least three iPhone cameras this way with my KTM 690.

Correction: random.randint + string interpolation

I bet, that this is mostly random.randint benchmark and not SQLite.

Also interesting whether batched numpy version would compare better to Rust.

Not yet, but Duplos and Technic support is widely requested, we will add support for them

Android app is under development right now.

One of the founders got sick of looking for Lego pieces while playing with his son. And he thought that there should be an app for that :)

So true!

We were hoping to start marketing alongside with the release of Android app )

Yes, Android app is being actively developed right now.

iPhone 6 would be very slow for a pleasant experience. If I'm not mistaken.

We use iPhone 8 as a "low end device" for testing.

Latest iPhones are performing extremely well on this task

Technically it can be done and we had prototypes that did something similar.

Hopefully mobile team will manage to implement all the improvements that come from users :)

We ignore color by default now.

MOCs library is not very big at the moment, but we did not intend to go viral either :)

The library of different builds will grow for sure!

> Showing me where pieces are located is the best part IMO as I’d usually spend most time looking for specific parts when building something.

Exactly that pain brought Brickit to life!

Not yet, but it seems that we will talk about it

To dig deeper, there are couple of interesting nuances there:

- extreme number of objects on single photo, typical number of visible pieces in large pile is 1500-2000

- extreme number of classes in multi-class classification, there are ~1000 most common Lego bricks and up to 30000 classes if you include rare bricks and different patterns

- really hard data labelling: one photo can take up to a 5 work days to label

That's what Lego does.

It seems almost intentional that each set has several rare pieces that can not be found in other sets.

I see. It would make sense to self-host this video.

Will forward it to product guys

Several neural networks, lots of hard work in data labelling, more hard work in data synthesis, even more hard work in dataset curation.. et voila!

Source: it was me with the team who did all of that.

You do not need any login to scan.

Just download the app and core functionality is ready to be used.

For completeness sake, I would like to mention, that Apache Drill works with local files and can be used for exactly the same purpose.

Do I understand correctly, that largest quantum computer that exists today contains less than 100 qubits?

Also it does not seem that there's an exponential grow in this area: https://www.statista.com/statistics/993634/quantum-computers....

They hit the wall in 2017.

We should be safe for now :)

It seems like you can substitute Haskell to Rust in almost every point they make.

(Or with many other great options)

I did not, thank you for the tip!

Thanks!

That's not exactly what I was looking for, though. Because every listed approach injects technical complexity in the middle of my business logic.

I.e. if I have two consequent tasks I have to define them as a separate scripts or commands, package them, upload and then orchestrate them in a completely different place.

While in Prefect I have all the niceness of writing almost plain Python (as with new tasks API in Airflow), then I can package and distribute the whole thing in docker image with a single command. It really matters!

Are there any plans for DAGs packaging in docker containers similar to what Prefect does?

Would be perfect to have separate dependencies for different DAGs, otherwise we always end up with a pile of everything ever needed with no clear way to remove obsolete packages from setup.

There’s an option for that: GIT_LFS_SKIP_SMUDGE=1 git clone SERVER-REPOSITORY

> The code is really just a pythonized version of the dot format.

Not exactly. By utilizing Python as a base language you gain access to control structures as `for` loops, `if`-s and functions. It can make your diagram script much more powerful and/or more complicated.

Postgresql is quite good easy to integrate alternative to Elastic.

It has built-in fulltext search: https://www.postgresql.org/docs/12/textsearch.html

> At the end of the day, I cant run ANY accelerated workloads using Intel graphics or AMD ....because there's simply no software support anywhere.

You can, there’s an Intel framework: OpenVINO that is targeting Intel hardware, processors and HD video cards, you can convert TF graph to Vino and use their inference server, that mimics TF serving.

There’s TF ROCm port by AMD as well.