liuliu

7 ms·

91.
by liuliu 8mo ago
I agree. It is either ads, or Anthropic way (which is: you are too poor to use our ChatBot). There is no other way to pay the > $1 trillion per year CapEx for building these chat bots. Would there be other way? Sure, it could be governme
92.
by liuliu 8mo ago
Why MLX doesn't just detect apple10 support (for Metal)? That excludes all the devices without NA.
93.
by liuliu 9mo ago
Collaborative software development is a high-trust activity. It simply doesn't work in low-trust environment. This is not an issue with code review, it is an issue with maintaining a trust environment for collaboration.
94.
by liuliu 9mo ago
Are you sure it is choked on writes not on reads and writes? SQLite default setup is inefficient in many ways (as well as it's default compilation options), and that often cause issues. (I am just asking: are you sure WAL is on?)
95.
by liuliu 9mo ago
I wish there are something for SwiftUI on Windows. I meant to support Windows for Draw Things, but the opportunity cost is too high without proper UI tooling.
96.
by liuliu 9mo ago
No need for the sarcasm. I am extremely generous about what FHE can achieve. Of course it is not 100x right now.
97.
by liuliu 9mo ago
> I think the US could seriously pull out of NATO and leave the EU to fend off Russia by itself. It'll have to start spending enormous tax dollars on defense and war. Disagree. If US pulls out of NATO, most likely scenario is EU con
98.
by liuliu 9mo ago
In theory, you trust the "crowd" (rather than the hosting entity) because if they don't do what they said, the "crowd" should make a noise about it and you would know.
99.
by liuliu 9mo ago
FHE is impossible. You cannot expect to compete on 100x more cost for the same service you provide (and there is no design for accelerated hardware (Tensor Core) on FHE).
100.
by liuliu 9mo ago
I agree it is more like e2teee, but I think there is really no alternative beyond TEE + anonymization. Privacy people want it locally, but it is 5 to 10 years away (or never, if the current economics works, there is no need to reverse the t
101.
by liuliu 9mo ago
I think there is only so much you can do practically. Without a secure "enclave", there isn't really much you can do. What's your alternative?
102.
by liuliu 9mo ago
One thing useful for Swift is it's native interop with C / C++ libraries. These are often presented as SwiftPM or Bazel dependencies. How do you handle SwiftPM dependencies?
103.
by liuliu 9mo ago
This is a common misunderstanding from industry observers (not industry practitioners). Each generation of (NVIDIA) GPU is an ASIC with different ISA etc. Bitcoin mining simply was not important enough (last year, only $23B Bitcoin mined in
104.
by liuliu 10mo ago
rules_oci (and bunch of rules_* under bazelbuild / bazel-contrib org on GitHub) is Bazel recommeded rule sets. I don't agree with your parent comment about Bazel, but your comment is not fair too. Bazel tries to be better build to
105.
by liuliu 10mo ago
I hope you finish this one though. It starts strong (I particularly liked how you looked into ncu and shows what each recommendation means, this is very helpful for beginners), but ends with something not satisfying. You didn't explore
106.
by liuliu 10mo ago
You can run newer models like Z Image Turbo or FLUX.2 [dev] using Draw Things with no effort too.
107.
by liuliu 10mo ago
They were talking about neural accelerators (a silicon piece on GPU): https://releases.drawthings.ai/p/metal-flashattention-v25-w-...
108.
by liuliu 10mo ago
I know this sounds weird: "symbolic execution" of pickle VM cannot be slow right? We are talking about just a few thousands instructions here and you don't need "symbolic execution" per se, just write a custom inter
109.
by liuliu 10mo ago
Agree. Writing a pickle interpreter is not particularly challenging. I did that in Swift to help load PyTorch checkpoint https://github.com/liuliu/swift-fickling without these pitfalls.
110.
by liuliu 10mo ago
> You can hand out chunks of sequential ids from a central coordinator to avoid collision; this is a well-established pattern. The problem is: is that part of postgresql? If not, someone has to write the buggy code for that well-establis
111.
by liuliu 10mo ago
I read it (and regret it is a waste of my time). Their arguments are: * integer keys are faster; * uuidv7 keys are faster; * if you want obfuscated keys, using integer and do some your own obfuscation (!!!). I can get on-board of uuidv7 (wi
112.
by liuliu 10mo ago
Oh! That's great to hear. Congrats! Now, I want to get the all-to-all primitives ready in s4nnc...
113.
by liuliu 10mo ago
I usually call it "head parallelism" (which is a type of tensor parallelism, but paralllelize for small clusters, and specific to attention). That is what you described: sharding input tensor by number of heads and send to respect
114.
by liuliu 10mo ago
But that's only for prefilling right? Or is it beneficial for decoding too (I guess you can do KV lookup on shards, not sure how much speed-up that will be though).
115.
by liuliu 10mo ago
They don't need to check outside payment links, until recently (I doubt they do though).
116.
by liuliu 10mo ago
Not saying M1 Ultra is great. But you should only get ~8x slow down with proper implementation (such as Draw Things upcoming implementation for Z Image). It should be 2~3 sec per step. On M5 iPad, it is ~6s per step.
117.
by liuliu 10mo ago
You are saying that you can raise $7b debt at double-digit interest rate. I am doubtful. While $7b is not a big number, the Madoff scam is only ~$70b in total over many years.
118.
by liuliu 11mo ago
There is no reason to believe Gemini Image is not diffusion model. In fact, generated result suggests it at least have VAE and very likely is a diffusion model variant. (Most likely a transfusion model).
119.
by liuliu 11mo ago
As a startup, they pivoted and focused on image models (they are model providers, and image models often have more use cases than video models, not to mention they continue to have bigger image dataset moat, not video).
120.
by liuliu 11mo ago
Windows 7 is the last good one. And that is only... Oh, almost 20 years ago. Never mind.

More ›