fpgaminer

11 ms·

61.
The Gory Details of Finetuning SDXL for 30M Samples(old.reddit.com)
1 points by fpgaminer 2y ago|0 comments
62.
by fpgaminer 2y ago
> At the end of 2022, there were at least 947 GW of utility-scale solar power capacity within the interconnection queues across the nation, 456 GW of which include batteries. Seems more likely to me that the utilities are expecting a lar
63.
by fpgaminer 2y ago
> I actually think StackExchange is quite well designed to optimize for good questions and answers. There's questions and answers on SE? I thought it was just pop-ups and sidebars? (I'm being mildly sarcastic, but I think this
64.
by fpgaminer 3y ago
My completely unexpert opinion, informed by listening to all the episodes of Darknet Diaries, agrees with this. US intelligence likes to just bully/bribe/blackmail the supply chain. They've got crypto chops, but I don'
65.
by fpgaminer 3y ago
> It can be less malleable than the older model and harder to work with to achieve a given desired result. That has been my experience as well. It's frustrating because SDXL can be exquisite, but SD 1.5 is more "fun" to w
66.
by fpgaminer 3y ago
Yeah, seems like the downsides are: * 3% explosive yield * Likely a dirty explosion, spreading radioactive contaminants. * The area where the beam is targeted (couple meter radius) gets a 1 Sv/sec dose, for about 100 seconds. "com
67.
by fpgaminer 3y ago
Same; a good ML focused discord would be great. Training ViTs all day is lonely work. I'm mostly locked into skimming the "Research" channels of image generation discords. LAION used to be decent with a good amount of inte
68.
by fpgaminer 3y ago
Seems like a solid paper from a skim through it. My rough summary: The popular large scale diffusion models like StableDiffusion are CNN based at their heart, with attention layers sprinkled throughout. This paper builds on recent researc
69.
by fpgaminer 3y ago
> A Vanguard spokesperson told CoinDesk that "spot Bitcoin ETFs will not be available for purchase on the Vanguard platform" and that it has no plans to offer Vanguard Bitcoin ETFs or other crypto-related products. The spokespe
70.
by fpgaminer 3y ago
How is that different from humans who prefer tools they know to tools they don't?
71.
by fpgaminer 3y ago
If comparing apples to apples, the 4090 needs to clock up and consume about 450 W to match the A100 at 350W. Part of that is due to being able to run larger batches on the A100, which gives it an additional performance edge, but yes in gen
72.
by fpgaminer 3y ago
Everything in moderation, including laws. We should view full legalization of "hazardous" activities in the same light as full prohibition. They are both black and white measures that increase harm. Somewhere in the middle, leg
73.
by fpgaminer 3y ago
The link is to news of a proposal, not a rule.
74.
by fpgaminer 3y ago
Somewhat tangential, but I hadn't heard about the Emu model, which was apparently released (the paper [1] at least) in September. I was curious about the details and read the Emu paper and ... I feel like I'm taking crazy pills r
75.
by fpgaminer 3y ago
> Public money indirectly comes from the most profitable companies (In the U.S.) Corporations only account for 6% of federal tax revenue ( https://taxfoundation.org/data/all/federal/us-tax-revenue-by... ). A
76.
by fpgaminer 3y ago
The architecture is quite compelling. I would not have expected it to work as well as it does. Glancing at the benchmarks it's basically on par with other VLMs in its class, despite having no separate image encoder. Is there an assoc
77.
by fpgaminer 3y ago
That was the first time I'd read about it on HN, but as pointed out on that HN post it wasn't the first time Softmax + 1 was proposed. And, AFAIK, it has never resulted in better performance in practice. Maybe Softmax + 1 works
78.
by fpgaminer 3y ago
Correct, but to be fair to readers (like me) the use of the term "infinite-length inputs" is misleading. Still, really interesting work. The most salient bit is the discovery shown in Figure 2, summarized as: > (1) The attenti
79.
by fpgaminer 3y ago
At least AWS gives me an invoice each month with a break down of fees. GCP just sends an invoice for the total amount. No detail, no breakdown. And I have never gotten any of GCP's many different pages under their billing system to
80.
by fpgaminer 3y ago
> which apply to all "businesses." It only applies to businesses that make over $25 mil, or that are in the business of selling user data. ( https://en.wikipedia.org/wiki/California_Consumer_Privacy_Ac... )
81.
by fpgaminer 3y ago
I've been quite pleased with California and the CCPA thus far. I've submitted a few deletion requests a now, and despite my jadedness all but one went through without a hitch. I reported the one to the California AG and within t
82.
by fpgaminer 3y ago
Delrin? You can machine it like steel and it has great temperature resistance. In fact, it gets used in modern automotive quite a bit.
83.
by fpgaminer 3y ago
I think most people should wait, yes. There's a HUGE amount of pressure to get these things approved for weightloss. So most people can just wait and watch a year or two. But to be clear, these drugs are already approved by the FDA f
84.
by fpgaminer 3y ago
If I may take a guess, your previous comment may be getting downvoted because people misread it as fat-shaming the U.S. (which is one of the world's leaders when it comes to being overweight). That's definitely how I read it the
85.
by fpgaminer 3y ago
Perhaps a joke, but it's a legitimate concern of a very complex topic. U.S. society has gone through a golden era of accepting that weight is often outside of people's control. All of us are addicted to something, it's just
86.
by fpgaminer 3y ago
SSRIs being over prescribe is a dangerous myth that should not be repeated; it puts people at risk. The vast majority of people get on SSRIs as a last resort to save their lives.
87.
by fpgaminer 3y ago
Llama off the top of my head: https://arxiv.org/pdf/2302.13971.pdf
88.
by fpgaminer 3y ago
I doubt it. If anything, ULMFiT era AI has finally killed the need for human curated data. ChatGPT 4 is already being used as an oracle model that everyday AI models are trained off of. A truly gargantuan oracle model will obviate all bu
89.
by fpgaminer 3y ago
That's effectively what RLHF is; a means for LLMs to self train on their own output exclusively by using a small human curated dataset as guidance as to what a "good" and "bad" output is.
90.
by fpgaminer 3y ago
> Was this not sort of the clear implication of the fact that most LLMs are currently only being trained with one epoch? Slight nit: Many public LLMs are trained for at least slightly over one epoch, and usually several epochs on particu

More ›