spott

19 ms·

61.
by spott 1y ago
Only if you know the data distribution. It is pretty easy to get 99.99% accuracy on a dataset that is 99.99% a single class for example.
62.
by spott 1y ago
I mean, how much money are they throwing at this? I doubt it approaches anything close to a percent of the cash they have on hand.
63.
by spott 1y ago
How much time does it take you to craft the PR compared to writing the code? Just curious if crafting the PR becomes significantly easier with practice.
64.
by spott 1y ago
What does it use for the graph part? Elasticsearch? A Postgres plugin?
65.
by spott 1y ago
> The upside for the user is to have a larger variety of devices to choose from, each with similar interoperability with his Apple device. That is the dream of the dma. It has not been proven to be the reality. The reality could very
66.
by spott 1y ago
The key here is in the cache lines. Sorting (really scanning through an array) is very efficient in cache lines, while hash tables are very inefficient. In the example, every memory access in hashing gets 8 byte of useful data, while every
67.
by spott 1y ago
How are you using it with Claude code?
68.
by spott 1y ago
I have a wireguard server at home, and it is great. The iPhone and Mac official wireguard clients allow you to set “excluded” wireless SSIDs, so if you are out and about, and not on a SSID that matches one of your excluded SSIDs, you are au
69.
by spott 1y ago
vLLM is a LLM model serving framework written using raw PyTorch. ONNX doesn’t support a bunch of operations that PyTorch does (it isn’t always possible to convert a PyTorch model to ONNX). Torchserve runs raw PyTorch. Generally speaking, Py
70.
by spott 1y ago
That is including all us military personnel, which puts it into perspective a bit.
71.
by spott 1y ago
I think this depends on how they are used. If you are leaning on the package manager for managing things like Python, then they are really annoying. If you are just skipping that and using something like UV, then you won’t care that LTS onl
72.
by spott 1y ago
Extraterrestrial in this sense is “didn’t come from earth”, not “aliens!”.
73.
by spott 1y ago
They are equivalent. In function signatures (what your article is talking about), using dict() instead of {} will have the same effect. The only difference is that {} is a literal of an empty dict, and dict is a name bound to the builtin
74.
by spott 1y ago
The propeller 2 is going to be used as the south bridge for the Ashet.
75.
by spott 1y ago
This depends on what you are trying to figure out. If you are talking about cosmology? Yea, you can look at existing data in new ways, cause you probably have enough data to do that safely. If you are looking at human psychology? Looking at
76.
by spott 1y ago
KV cache for dense models is order 50% of parameters. For sparse moe models it can be significantly smaller I believe, but I don’t think it is measured in kb.
77.
by spott 1y ago
Yea, I figured that out after I finished the article. The dangers of posting before I finish the article.
78.
by spott 1y ago
Personal theory: the rise of ETFs as a fallback means that money isn’t leaving the stock market anymore. Instead of people selling off and going to cash, they go to SPY, which doesn’t have the downward pressure on the stock market that go
79.
by spott 1y ago
All the Llamas have done it (well, 2 and 3, and I believe 1, I don't know about 4). I think they have a citation for it, though it might just be the RoPE paper ( https://arxiv.org/abs/2104.09864 ). I'm not act
80.
by spott 1y ago
It is interesting that openai isn't offering any inference for these models.
81.
by spott 1y ago
Groq is offering 1k tokens per second for the 20B model. You are unlikely to match groq on off the shelf hardware as far as I'm aware.
82.
by spott 1y ago
Why is just capturing real values not the answer here? Something like grab .01% of the inputs to that function for a day or something like that (maybe a lower fraction over a week or month). Is the cost of grabbing this that high?
83.
by spott 1y ago
Do you know what this means in practice? Is it that I have to fall back to git for its changes? Or that I just shouldn’t use jj if I’m in a repo with lfs files?
84.
by spott 1y ago
My reading of this wasn’t that Apple has a bunch of security bug fixes they aren’t upstreaming, it is that they are maintaining their own forks of an old version and back porting security bug fixes from upstream into their fork. Maybe they
85.
by spott 1y ago
This makes an assumption that a bunch of companies are maintaining their own forks of MIT software with bug fixes and features and not giving it back. I find that hard to believe.
86.
by spott 1y ago
But not in the browser… because it depends on node.js apis.
87.
by spott 1y ago
I believe it is a majority vote kinda thing, rather than a best single result.
88.
by spott 1y ago
Does anyone know of any local models that are capable of the same tool use (specifically web searches) during reasoning that the foundation models are? I realize they aren’t going to be as good… but the whole search during reasoning is pret
89.
by spott 1y ago
From March 2024
90.
by spott 1y ago
That is an expensive dev board for something like this… Here is a esp32-h2 with thread and zigbee support for $5-$12 https://a.aliexpress.com/_mMfZEsX I figure support for something like this is likely to increase faster th

More ›