• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

uOpt's latest activity

  • U
    uOpt replied to the thread Nvidia's Project DIGITS.
    That's what I have been using so far, with a 5090. But I want an additional runner for larger models with larger context sizes.
  • U
    uOpt reacted to mashie's post in the thread What LLM/Model you using? with Like Like.
    It is a smart way to keep KV caches reasonably small at large contexts (1M). According to the friendly neighbourhood LLM: Hybrid attention in Large Language Models (LLMs) is an architectural approach that combines traditional full (softmax)...
  • U
    uOpt replied to the thread What LLM/Model you using?.
    What is Hybrid Attention? Both of these cards mention it.
  • U
    uOpt replied to the thread Nvidia's Project DIGITS.
    Maybe somebody here can explain this to me: I have been told that not all LLMs "fit" the Apple metal backend. From what I know about the metal api (and the requirement to use it instead of programming hardware directly) and how much difficulty...
  • U
    uOpt replied to the thread AI LLM hardware discussion thread.
    The Apple Studio lease program looks more attractive than not. I would add a solid $224 tax per month to my expenses.
  • U
    uOpt replied to the thread AI LLM hardware discussion thread.
    New Mac Studio with M5 Ultra can be preordered. 256 GB max for now, 512 GB in October. https://www.apple.com/mac/compare/?modelList=Mac-mini-M6,Mac-mini-M5-Pro,Mac-Studio-M5-Ultra...
  • U
    uOpt reacted to daglesj's post in the thread Recent experiences with Headphones / Computer Audio with Like Like.
    Recently got some Audioengine HD3 speakers and slapped a Topping E30 DAC into it with some silver RCA phonos I have had for years (it was those or some Furukawa FD phono RCA cables that cost like £250 back in 1992). Rather lovely little combo.
  • U
    uOpt replied to the thread What LLM/Model you using?.
    I wanted to give Qwen 3.8 another try. As a reminder, on my standard code review benchmark is it reasoning for 45000 tokens, which at my 55 t/s is just too slow. So I restricted thinking with "--reasoning-budget 4096". But then it doesn't give...
  • U
    uOpt replied to the thread What LLM/Model you using?.
    I just went back from Qwen 3.8 to 3.6 because it is just too thinking-happy and hence too slow.
  • U
    uOpt reacted to Mad Maxx's post in the thread Pixel 11 / Pixel 11 Pro / Pixel 11 Pro XL / Pixel 11 Pro Fold with Like Like.
    Delivered today, will setup on Saturday.
  • U
    uOpt reacted to FRZ's post in the thread Pixel 11 / Pixel 11 Pro / Pixel 11 Pro XL / Pixel 11 Pro Fold with Like Like.
    Lighting isn't very good but here is a couple pics
  • U
    uOpt posted the thread ST:TNG on Paramount+: volume fluctuations lately? You, too? in General Software.
    I just restarted binging ST:TNG on Paramount+ in the US, english audio. In the episodes I watched (beginning of 3rd season) there are several sharp audio volume changes in the middle of episodes. It is possible that this is limited to the center...
  • U
    uOpt replied to the thread Intel will supply Raptor Lake for years.
    I have a 14900KS system with DDR4. It's a real road runner.
  • U
    uOpt replied to the thread What LLM/Model you using?.
    Yeah, Qwen 3.8 is overthinking things. It uses 4x as many output tokens for me than 3.6. But as you say, the results are excellent.
  • U
    uOpt reacted to pendragon1's post in the thread MacOS can't reach udate server anymore with Like Like.
    osxdaily posts them https://osxdaily.com/where-download-macos-installers/
  • U
    uOpt replied to the thread MacOS can't reach udate server anymore.
    That doesn't work either for the *.pkg, because it wants to "personalize" the update, which it does through the update server, which it can't reach. However, I found what was causing it. I noticed that although I could ping www.google.com I...
  • U
    uOpt replied to the thread MacOS can't reach udate server anymore.
    I thought the updates are not directly downloadable anymore since a few versions back? I have waited for a week after stopping the cronjob, I don't think it will reset itself.
  • U
    uOpt replied to the thread What LLM/Model you using?.
    If you use FreeBSD you get a whole bunch of AI tools such as hermes as proper OS packages.
  • U
    uOpt replied to the thread What LLM/Model you using?.
    The censoring is not in the learning. It is post-training. That is why you can remove the censoring (if you have an LLM to ask how :)). You just "rewind" so to speak.
  • U
    uOpt replied to the thread What LLM/Model you using?.
    Same code review: 3.6 uncensored: total time = 154571.82 ms / 12846 tokens 3.8 default: total time = 844671.49 ms / 61232 tokens
Show older items
Back
Top