gunalx

8 ms·

181.
by gunalx 1y ago
But they also squesed a 80% cut in O3 at some point, supposedly purely on inference or infra optimization
182.
by gunalx 1y ago
The system as bad as it is can be good as well. Let some consultants find out how to get the bids, and others how to do it well, and can be a outsourced third party. It gives differentiation without everyone needing to be large enough to a
183.
by gunalx 1y ago
The problem is more that they dint go the OSI approved license with attribution clause, and forced that into a BSD license.
184.
by gunalx 1y ago
But the 120b model has just as bad if not worse formatting issues, compared to the 20b one. For simple refactorings, or chatting about possible solutions i actually feel teh 20b halucinates less than the 120b, even if it is less competent.
185.
by gunalx 1y ago
Dosent work in firefox, because of missing webgpu.
186.
by gunalx 1y ago
I have in many cases had better results with the 20b model, over the 120b model. Mostly because it is faster and I can iterate prompts quicker to choerce it to follow instructions.
187.
by gunalx 1y ago
Should be doable yes.
188.
by gunalx 1y ago
This is real good also wfor installing win11, because it at least did the local user and requirements bypass patches last time i used it.
189.
by gunalx 1y ago
You probably aren't working that hour. (also way better than 8).
190.
by gunalx 1y ago
You say it is always empty, but. I have seen weird issues coming from /dev/null not actually being empty but being a file or symlink (dont remember) With garbage data.
191.
by gunalx 1y ago
Where I live we are already talking about abandoning 2g, and 4g. (3g died a while back and 2g was kept as legacy network).
192.
by gunalx 1y ago
google tpm, groq and cerebras needs yo be mentioned even if they are more general architecture optimized.
193.
by gunalx 1y ago
Depends. API pricing from oss model inference providers basically has to be sustainable, because of competition in the space. And with that in mind, i definetly dont use more than a couple of bucks a month in API refils. (not that i really
194.
by gunalx 1y ago
It is soo cool. Had the opportunity to see a demo at last year fossdem, and it blew me away with the level of configurability and seemingly compatibility.
195.
by gunalx 1y ago
My experience is exactly opposite. Claude excelling in ui, and react. While gpt5 being better on really niche stuff, migth just be me better at caching when gpt5 halucinates as opposed to the claude4 hallucinations. But after openai started
196.
by gunalx 1y ago
So significantly worse than qwen2.5, kinda useless in the current landscape. but always fun with more arcitechtures.
197.
by gunalx 1y ago
you might habe an old enough rm account to have free base subscription.
198.
by gunalx 1y ago
not nessesarily, but you'll have an issue if they do a controll, or it comes up.
199.
by gunalx 1y ago
vscode has gotten way to bloated in recent years in my opinion. zed is just a really fresh Beath of air.
200.
by gunalx 1y ago
It should have been properly setup, but it was a couple years ago and I migth have left to much open. (it was on a VM behind a consumer nat, firewall router solution )
201.
by gunalx 1y ago
ist that the whole point of unifying recovers as well? or am I missing something?
202.
by gunalx 1y ago
This really seems like they did everything they could and still got abused by borderline criminal activity from scrapers. But i do really think it had an impact on scraping, it is just a matter of attrition and raising the cost so it shoul
203.
by gunalx 1y ago
You would still need drivers and all the stuff difficult with nvidia in linux with a egpu. (Its not nessecarily terrible just suboptimal) Rather just add the second GPU in the Workstation, or just run the llm in your AMD GPU.
204.
by gunalx 1y ago
Ran archive warrior a while back but hadde to shut it down AS i sterted seeing the VM was compromised trying to spam ssh and other login attemps in my local network.
205.
by gunalx 1y ago
Maybe using a larger model to generate synthetic data of question path Combos, and also to rephrase and generate similar type questions for a more varier training set.
206.
by gunalx 1y ago
Im im the same boat. Luckily in ny case a local banking app has their own NFC card Funktion witch works flawlessly. But no tap to pay would for me have been one of the greatest downsides with graphene os.
207.
by gunalx 1y ago
Kagi seems to partially be that. Yes really corpo but way Better wibes than Google. Searxng is a bit diffrent but also a thing.
208.
by gunalx 1y ago
Mistral also has small medium and large. With both small and medium håving a thinking one, devstral codestral ++ Not really that mich simpler.
209.
by gunalx 1y ago
Would love to se something like this trained for multilingual purposes. It seems kinda like the same tier as piper, but a bit faster.
210.
by gunalx 1y ago
It is in the name, rules for the robots. Any scraping ai or not, and even mass recrsive or single page, should abide by the rules.

More ›