zwaps

27 ms·

181.
by zwaps 2y ago
I mean this is implicit in back propagation, say, you need to store gradients anyway but if you get to a zero loss than you are just done.
182.
by zwaps 2y ago
Is it me or is this paper basically missing all technical information? I get that Therese proprietary technology, but if so, can we please not put this on arxiv and pretend it’s a scientific contribution?
183.
by zwaps 3y ago
It should be mentioned that the original author Nils Reimers has moved on and the repo has been stale since 2021/2022. It has recently (e.g. end of 2023) gotten a new team and has since seen updates. This is obviously quite significant
184.
by zwaps 3y ago
I am confused Brutalist Web Design advocates for some simple rules which this thing doesn't seem to follow. It has more Javascript and the back button doesn't work. So, compared to this, the original Hacker News is actually Brutal
185.
by zwaps 3y ago
sadly, w&B means you have to upload it to the cloud, which is not possible in every case :(
186.
by zwaps 3y ago
Flan-T5 is actually not a bad model, even today (for fine-tuning)
187.
by zwaps 3y ago
email grab at the end tho
188.
by zwaps 3y ago
the latter would not be very smart seeing that Langflow is pretty much one of like three viable alternatives for what it does in a potentially giganormous market
189.
by zwaps 3y ago
"Tip: Commercial and non-commercial users of Redict are not required to publish or otherwise “open source” the source code of Redict, including any private modifications they make to the source code" And very next section: "I
190.
by zwaps 3y ago
If you have a commercial use-case, there is also a non copyleft fork available here https://www.linuxfoundation.org/press/linux-foundation-launc...
191.
by zwaps 3y ago
What I find interesting here is how far a well working QA application with LLMs (such as this one) is away from anything that can be generalized to other topics. Thats probably where we are right now: I have seen quite a few purpose built a
192.
by zwaps 3y ago
1. They compare with an older sort of standard implementation of a transformer Unsure whether the results would be equally significant compared to models with gated units or multiquery etc. 2. The difference seems to diminish with scale. Re
193.
by zwaps 3y ago
Fnn are easy for sequential data because there is no relationship between the elements. Hence, these NN are a frequent target of simplified analyses. However, they also never led to exciting models which we now call AI. Instead, real models
194.
by zwaps 3y ago
Before I read this, is this yet another paper where physicists believe that modern NN are trained by gradient descent and consist of only linear FNN (or another simplification) and can therefore be explained by <insert random grad level
195.
by zwaps 3y ago
Got it, so the underlying infrastructure (the inference nodes, if you wish) would be something to be solved outside of Hatched, but it would then allow to schedule inference tasks per user with limits.
196.
by zwaps 3y ago
You say this is for generative AI. How do you distribute inference across workers? Can one use just any protocol and how does this work together with the queue and fault tolerance? Could not find any specifics on generative AI in your docs.
197.
by zwaps 3y ago
Depends on country. In the north, no way.
198.
by zwaps 3y ago
What a terrible article
199.
by zwaps 3y ago
That's not a lot compared to 8xA100
200.
by zwaps 3y ago
Also all Stat Trek Federation ships
201.
by zwaps 3y ago
It really isn’t
202.
by zwaps 3y ago
I bought a washing machine from them and it had continuous problems for several months, until I got a refund. The technician advised me to buy from Miele, Siemens or Bosch, as Samsung apparently has lots of issues.
203.
by zwaps 3y ago
Most economists (who write these sort of textbooks) have some sort of math background. The push to find the most general "math" setting has been an ongoing topic since the 50's and so you can probably find what you are lookin
204.
by zwaps 3y ago
I think everyone who comes from a different literature where academic "rigor" is higher and similar results already exist (in the author's case, he is aware of Kernel results) is infuriated by ML papers like "Attention i
205.
by zwaps 3y ago
My consulting company bills me out at a much higher multiple than that in my gross salary
206.
by zwaps 3y ago
What is a Langchain like Vector db? Langchain is not a vector db?
207.
by zwaps 3y ago
It was trained on this book so again, this is not a good test It will know the answer even without the book
208.
by zwaps 3y ago
This needs to be shown. For example, asking for something that is clearly in the training data (like Paul Grahams cv) is certainly not a proper way to test context recall
209.
by zwaps 3y ago
Excellent article. This suggests the Gpt scalings are like Rope scalings and one should not go beyond 2x original context length. If Claude2 has an internal Rag, then this means also that the 200k context length only holds for queries that
210.
by zwaps 3y ago
Hard disagree, this is a play from the top

More ›