emu

https://x.com/jacobaustin132/status/1886844716446007300 is the announcement of this book.

The uncertainty alone is already stopping people from making job offers this year in physics. Will the department of Energy keep funding fundamental physics research, for example? Will the NSF or NASA? No one knows. The rational response under these circumstances is to pull back on hiring.

Such capability can’t even be contemplated on the TPU side because you basically get a black box.

I'll just leave this here: https://jax.readthedocs.io/en/latest/pallas/index.html

Baratza also sell replacement parts for their grinders on their web site, and provide clear instructions on how to install them. I am delighted to support a manufacturer that builds repairable products.

But the interpretation of the journalist is the issue here. Comparing a cost optimized part (TPU v5e) against a performance optimized part (H100) and deciding that this makes Google "behind" is just incorrect.

No, it is not. That's the sparse fp8 flop number, but you need to ignore sparsity and compare bf16 flops not fp8 flops for the comparison the ancestor post is making.

Yes, we made this more formally supported recently.

We felt that Windows CPU support was important so everyone can run JAX, even if it's not always the most-accelerated version of JAX. And we got some great PRs from the community that helped fix a few open issues.

We ship Windows CPU only at the moment.

We don't support Windows GPU because we haven't had the engineer bandwidth to support it well.

We recommend WSL2 for GPU on Windows at the moment because that is a compromise: it allows CUDA support, without us having to support another release variant.

But we welcome community contributions!

Not true!

We release Windows CPU wheels (https://pypi.org/project/jaxlib/#files). So JAX on CPU works great on Windows.

We don't release Windows GPU wheels at the moment, but that's because we're a small team and none of us use Windows personally. We welcome contributions!

(I verified that the Windows CUDA GPU support built as recently as two weeks ago, but I don't have the ability to test that it works.)

We recommend WSL2 because that's just using our existing Linux CUDA release.

If pypi didn't enforce size limits, I'd agree with you. We routinely have to delete older versions of our package to make space for newer versions.

I enjoyed playing Stellaris on my MacBook M1 pro on a transcontinental flight last night. 4x games work well for passing the time!

I wish Mac OS and Apple hardware supported 8k monitors. I have this monitor and love it, but I can't use it with a Macbook at higher than 4k.

My employer withheld California tax on restricted stock grants for me exactly as described here when I moved out of California years ago.

Other states even have forms for you to submit this calculation to them, see for example New York's IT203-F Schedule B. But you still have to apportion the income between states like this whether there's a form to do it on or not.

Yes, the difference between 4k and 8k text is easily noticeable for a 32" monitor, and I'm not particularly eagle-eyed. Text is sharper and finer details are better rendered. For videos and images I don't notice the difference in practice.

I had a 4k 32" monitor at work and found that it simply didn't give sharp, high resolution text, driven by either a Macbook or by a Linux box. And you wouldn't really expect that, either: a 4k 32" monitor is only ~140dpi, which is only marginally higher resolution than the ~100dpi screens we had for many years.

I think the best point of comparison for the Dell UP3218k monitor which I have is a Retina Macbook Pro screen: subjectively, it's a similar experience in terms of text sharpness and legibility (>200 DPI, glossy), just in a 32" form factor.

I suspect 8k at 32" is actually a bit higher resolution than necessary (~280dpi), but there's nothing else on the >30" high resolution monitor market other than Apple's 6k display, which is significantly more expensive.

Be aware that there's no Mac OS support for 8k displays, but Linux and Windows on a desktop with a reasonably modern NVidia GPU work great.

I love using a 32" 8k monitor for working with text/code. To me, as a software engineer who stares at code all day, higher quality text rendering is sufficient justification for 8k.

I tried playing a few games at 8k (just for fun) and found that it simply wasn't worth the frame rate hit, or really even that noticeable an improvement with the assets in the games I tried.

One nice property of 8k resolutions is that they have an integer scaling factor from both 4k (2x) and 1440p (3x), so if you have an 8k monitor you can play games at either of those resolutions with high quality scaling.

We (JAX) see JAX and Numba as mostly complementary; they have different strengths and they are focused on different things.

We haven't tried combining them yet, but we think it would be fun to explore (https://github.com/google/jax/issues/1870). For example, you could use Numba to hand write a numerical kernel that then participates in a machine learning model that uses JAX automatic differentiation.

Good catch, I missed Trax! Trax is a configuration-driven neural network framework focused on sequence model research, as a successor to Tensor2Tensor.

Comparisons are hard in general and I don't have a good answer for you right now, but keep in mind most of these libraries are from researchers openly sharing the codebases they develop for their own work. We see the role of JAX as analogous to NumPy, that is, a common substrate on which folks can build these sorts of tools.

Hello from the JAX team!

We'd like to take this opportunity to give a shout out to some of the awesome projects folks are building on top of JAX, e.g.,

* Flax, a neural network library for JAX (https://github.com/google/flax)

* Haiku, a neural network library for JAX inspired by Sonnet (https://github.com/deepmind/dm-haiku)

* RLax, a library for building reinforcement learning agents (https://github.com/deepmind/rlax)

* NumPyro, a probabilistic programming library on top of JAX (https://github.com/pyro-ppl/numpyro)

* JAX-MD, a differentiable molecular dynamics package built on top of JAX (https://github.com/google/jax-md)

Autodiff is certainly one of the strengths of JAX. See the JAX Autodiff cookbook for a flavor of what JAX can do: https://colab.research.google.com/github/google/jax/blob/mas...

You might also like the per-example gradients example that appears first in the JAX Github page: this is only one line of code, but important for research areas such as differential privacy.

Since you mention it: Cloud TPUv3 is now available in beta (brand new!) See: https://cloud.google.com/tpu/

(420 TFlop/s, 128GB HBM)

If you look, the same listing has 5 fulfilled-by-amazon shippers as well :-)

The listing says, and I quote: "Apple A1385 USB Cube Adapter ..." (in the title) and "Includes 1 Apple certified charger cube". I'd bet that neither is true.

It's trivially easy to run into counterfeit products on Amazon. Here's an example I pulled up in 2 minutes of searching:

https://www.amazon.com/Apple-A1385-Adapter-Charger-iPhone/dp...

I see two sellers offering an iPhone charger, ostensibly genuine and new, as low as USD 3.19.

Compare with the Apple listing for the same item for USD 19: http://www.apple.com/shop/product/MD810LL/A/apple-5w-usb-pow...

There is no way you can be selling the genuine product at that price and be making a profit. I don't know what Apple's wholesale price for chargers is, but it is presumably more than USD 3.19.

Amazon has lost at least two orders from me recently because I had no confidence that I would receive a genuine product.

The documentation is now up too: https://www.tensorflow.org/versions/master/experimental/xla/

The code is up:

https://github.com/tensorflow/tensorflow/commit/1e67c90e2cac...

https://github.com/tensorflow/tensorflow/tree/master/tensorf...

The corresponding documentation hasn't been pushed yet; I'll post a link when it is up.

Note that XLA is work in progress --- we're releasing the code early because we want to get the community involved. The GPU backend is in good shape, and improving by the day. We haven't had as much time to devote to the CPU backend, and it only has limited support for parallelism. Contributions welcome!

Precisely. Just-in-time compilation allows the compiler to specialize the generated code for the shapes of the Tensors that appear at run time. This allows us to generate better code.

XLA also has an experimental ahead-of-time mode, which we think will be particularly interesting for some production and mobile deployments. This is all work in progress though, and we're looking forward to getting the community involved.

I'm on the Tensorflow team.

Unofficial answer: no promises, but it should be open-source soon. It may even be released in the next day or two. Watch this space!

Yes. I use the Dell display with a 2015 Macbook Pro, using dual Displayport cables. It works pretty well. I think you need the model with the AMD M370X GPU to drive an external 5k display.

I also use the Dell 5k monitor with a PC with a Geforce 980 GPU; it works surprisingly well in Windows 10. (You can even play some games at 5k ---- Civilization VI is really pretty at 5k!)

Linux support is woefully bad, however. I guess this will change as these HiDPI configurations become more common.

I'm a happy owner of a Dell UP2715K 27" 5k monitor, which is around 220ppi. At 2x scaling, I find it to be a very pleasant pixel density for working with code; text looks gorgeous.

My only complaint is I wish the monitor was a touch larger. I would be very interested in a 32" 8k display when the price drops a little bit; 4k 32" displays just don't have a high enough pixel density for my taste.

See https://www.youtube.com/watch?v=W71BTkUbdqE for a lot more details on what Google does.