liuliu

7 ms·

61.
by liuliu 6mo ago
It needs a mlx fork because the lowest bit in mlx is 2 currently (for affine quantization).
62.
by liuliu 6mo ago
Close enough to a humanoid then you can move to places that humans can move to / around.
63.
by liuliu 6mo ago
I think for that to use pretext is to join each row with hard line break and then do prepare once, then walk each line. At least that will put the single layout performance into the best light. I am skeptical getting row height of many item
64.
by liuliu 6mo ago
prepare uses measure text, if it is in a for loop, it won't be fast. This library is meant to do prepare once and then layout many times. layout calls should be sub-1 ms.
65.
Draw-things-CLI: local media generation from command-line on your Mac(releases.drawthings.ai)
3 points by liuliu 7mo ago|0 comments
66.
by liuliu 7mo ago
Still have 4 brand new ones in my storage unit. Just in case these moments. Joke aside (I do have them tho!), I don't think Optane is that much use (not to mention it is only 256GiB for my unit). It is useful legacy crutch if you have
67.
by liuliu 7mo ago
I don't think this argument is wrong. But also debatable. At the end of the day, we are talking about the manifold of the reality (as compressed by LLM through language abstraction). It is remain to be seen if supervised fine-tuning on
68.
by liuliu 7mo ago
The problem is that it cannot access your credentials hence useless.
69.
by liuliu 7mo ago
There are, look no further than jemalloc API surface itself: https://jemalloc.net/jemalloc.3.html One thing to call out: sdallocx integrates well with C++'s sized delete semantics: https://isocpp.org/fi
70.
2 Days to Ship: Codex-Authored Metal Compute Shaders in Draw Things(engineering.drawthings.ai)
1 points by liuliu 7mo ago|0 comments
71.
by liuliu 7mo ago
To clarify, GPL is not a free as in "free gift", but it is free as in "freedom". The giving back part is strongly related to the "freedom", not related to whether you profit from it or not.
72.
by liuliu 7mo ago
GPL is not for you to make money. It is for the end-users to have freedom with their hardware. If you want to make money, use a proper license. To expand on this, GPL is not against capitalism neither. Sometimes, end-users' freedom wit
73.
by liuliu 7mo ago
Not really. The U.S. can send in the ground force to restore the trade around the Gulf. The BUT is obvious in this case tho.
74.
by liuliu 7mo ago
This is not different from mlx-lm other than it uses a closed-source inference engine.
75.
by liuliu 7mo ago
> It'd be nice if Python std lib had more thread safe primitives/structures (compared to something like Java where there's tons of thread safe data structures) Hence why basic Python structures under free-threaded Python a
76.
by liuliu 7mo ago
It is not going to reduce your workload. It is going to remove one of your co-workers.
77.
by liuliu 7mo ago
apples v.s. oranges. The later is true, Emad did get sabotaged (for not being able to raise money in time, about 8-month before he's leaving). Junyang didn't have that long arc of incidents.
78.
by liuliu 7mo ago
As much as I wish, it is going the other way. Caring about the 3 requires literacy, which in the world of LLM, is one thing that going to be reduced as a whole for human-kind.
79.
by liuliu 7mo ago
> and Apple is the only company that stands to benefit from it. And that is exactly why it won't happen (like that).
80.
by liuliu 7mo ago
That's a good thing right? In a capitalist society, you cannot just burn $300B without consequences. Not to mention it is not just anyone's money. It is Saudi's.
81.
by liuliu 7mo ago
It may not be obvious. But this is actually a good thing when we looking back in a few years. I always feel weird that executive branch can just destroy private enterprise with "Supply-chain Risk" / "Terrorist List"
82.
by liuliu 8mo ago
Every business metrics needs people to safeguard. That's how you get the number of ppl.
83.
by liuliu 8mo ago
Depending on what you do. If you are doing token generations, compute-dense kernel optimization is less interesting (as, it is memory-bounded) than latency optimizations else where (data transfers, kernel invocations etc). And for these, Ma
84.
by liuliu 8mo ago
The bandwidth is free on Cloudflare R2. I paid money for storage (~10TiB storage of different models). If you only host 1GiB file there, you are only paying $0.01 per month I believe.
85.
by liuliu 8mo ago
It is very simple. Storage / bandwidth is not expensive. Residential bandwidth is. If you can convince people to install a bandwidth-related software on their residential homes, you can then charge other people $5 to $10 per 1GiB bandw
86.
by liuliu 8mo ago
> We have a local model we would like to distribute but don't have a good CDN. That is not true. I am serving models off Cloudflare R2. It is 1 petabyte per month in egress use and I basically pay peanuts (~$200 everything included)
87.
by liuliu 8mo ago
Sure. If you treat "guard model" as diversification strategy, it is another layer of protection, just like diversification in compilation helps solving the root of trust issue (Reflections on Trusting Trust). I am just generally s
88.
by liuliu 8mo ago
The solution is to make the model stronger so the malicious intents can be better distinguished (and no, it is not a guarantee, like many things in life). Sandbox is a basic, but as long as you give the model your credential, there isn&#x27
89.
by liuliu 8mo ago
Agree. I think it is just people have their own simplified mental model how it works. However, there is no reason to believe these simplified mental models are accurate (otherwise we will be here 20-year earlier with HMM models). The simple
90.
by liuliu 8mo ago
Note that Qwen Image 1.0 (2512) wasted ~8B weights on timestep embedding. Both Z-Image / FLUX.2 series corrected that.

More ›