- Ones with better answers. Twitter dumbs down grok.
- For my LG TV, it is as deep as you can possibly go! And I think it took two options to fully opt out (including turning off the microphone).
- Nice. I have a simple system for typescript [1] where you can string tasks together like:
import { Workflow } from "@workglow/task-graph"; const workflow = new Workflow(); workflow .DownloadModel({ model: ["onnx:Xenova/LaMini-Flan-T5-783M:q8", "Universal Sentence Encoder"], }) .TextEmbedding({ text: "The quick brown fox jumps over the lazy dog.", }); await workflow.run();
It automatically caches results of steps if you say `new Workflow({outputCache});`
PS: the example above uses a local onnx model (it switches on model name) which is a good candidate for caching, but can be anything.
You can play around writing this way if you open the console at the web example [2] which has fun console formatters not enough people know about or use in DevTools. You can write code in the console and the example rewrites it as a json format in the web page (and visa-versa).
Or just use the free web app with local user account and local models to play around with. [3]
[1] https://github.com/workglow-dev/workglow
- Do they mention tagging your Bluetooth IDs at the same time?
- It’s right there in your TV’s settings though. Personally, I don’t trust them to obey the setting so my TV has no internet and I use an Apple TV.
- onnx supports CoreML, is that how?
- I bet there is one piece of silicon per two ports.
- Maybe we find out why things work in mice and not us.
- “ Photo of a blond male in his 50s with half gray hair “
Still fails. Every photo of a man with half gray hair will have the other half black.
- I am using ReactFlow, which has been really nice.
- https://workglow.dev
It is a workflow graph automation site (drag and drop and connect nodes), but is a toy and only allows a local user and local models (both transformers.js and tensor-flow mediapipe), so costs me nothing. Mostly text stuff at the moment, but working on a slate of image stuff this week, may get to audio and video as well, we shall see.
- That is like the HTML specification -- nobody ever puts up a web page that is not conformant. ;p
The idea behind putting some time as prefix was for btree efficiency, but lots of people use client side generation and you can't trust it, and it should not matter because it is just an id not a way of registering time.
- They are inseparable for Apple. CPUS/GPUs/memory. They can use chipsets to tweak ratios, but I doubt they will change the underlying module format—everything together.
My suggestion is to accept that format and just provide a way to network them at a low level via pci or better.
- Really, that’s a place for the MacPro: slide in SoC with ram modules / blades. Put 4, 8, 16 Ultra chips in one machine.
- What is good at SVG design?
- Not getting permits, and no permits required are two different things.
Unless you got cash, then it’s the same.
- Lame on user machines, but sometimes needed in a server environment. Easier to detect if someone is hauling off with your database as that will be the one you can’t see what’s going on. Of course, solve one problem and introduce three more.
- Third parties with tech could fix it.
- I guess today, instead of 128bit pointers we have 64bit pointers and secret provenance data inside the cpu, at least on the most recent shipped iPhones and Macs.
In the end, I’m not sure that’s better, or maybe we should have had extra large pointers again (in that way back 32bit was so large we stuffed other stuff in there) like CHERI proposes (though I think it still has secret sidecar of data about the pointers).
Would love to Apple get closer to Cheri. They could make a big change as they are vertically integrated, though I think their Apple Silicon for Mac moment would have been the time.
I wonder what big pointers does to performance.
A local model running alone on your machine will 100% always return the exact same thing and the internal state will be exactly the same and you can checkpoint or cache that to avoid rerunning to that point.
But… conditions can be different, and batching requests tends to affect other items in flight. I believe Thinking Machines had an article about how to make a request deterministic again without performance going to complete crap.
I tend to think of things this way (completely not what happens though): what if you were to cache based on a tensor as the key? To generate a reasonably sized key what is an acceptable loss of precision to retrieve the same cache knowing that there is inherent jitter in the numbers of the tensor?
And then the ever so slight leak of information. But also multiplied since there are internal kv caches for tokens and blah blah blah.