WHOADIE
Region: US
Saturday 06 June 2026 15:39:22 GMT
325003
16907
246
1856
Music
Download
Comments
BotRaper :
This video is misleading. First of all, Google didnt make turboverc. They made TurboQuant, then a third-party made Turbovec. Second, Turbovec does NOT compress the model itself or its weights. Turbovec compresses its KV Cache, which can be useful for things like Retrieval Augmented Generation (RAG). You got me super excited and then you filled me with rage when I found the truth. Thanks a lot 😠
2026-06-09 03:33:30
142
Ocmer :
So can I buy RAM at reasonable prices yet?
2026-06-06 17:19:32
631
Donald duck :
raspberry pi ai ?
2026-06-06 16:31:31
27
That_new_guy :
It’s called “middle-out” checkout the weissman score, Nucleus tried to re-create it
2026-06-06 22:23:11
150
𝖀𝖗𝖎𝖊𝖑 ☬ :
I'm currently reading the book Abandoned Principles of Reality by Alon Flake and all I can say is wow. The style of writing is a unique experience
2026-06-07 13:13:34
325
Ooga Booga Chuckapoo :
So Can I load a 120B model on my 16GB Vram?
2026-06-07 08:00:46
35
ismail :
So a 24gb model can fit in 6gb vram?
2026-06-06 21:27:03
6
zispaganu :
We have Winrar typa thing for ai
2026-06-06 16:19:38
412
theoriginalbearjew :
This mirrors the first season of Silicon Valley. lol
2026-06-07 02:54:38
17
🇨🇦XiLLest Reaper🇨🇦 :
so if Google designed it why is your name on it lol
2026-06-06 22:38:22
14
Richard CoJones :
then why do we need so many data centers?
2026-06-06 18:38:03
7
Morkit254 :
why don't they shrink my Gmail storage. 😏
2026-06-07 05:45:02
8
Yamzz :
ai memory shrinking but it's model still as big as terabytes
2026-07-19 23:43:51
2
Msunu ka tyla :
how does even GitHub work? 😩
2026-06-08 11:45:18
0
MarcusGameDev :
why are we celebrating? the AI companies will still charge the same amounts, this only helps multi billionaires
2026-06-08 07:52:25
2
Freddy :
I dont think people understand were in an arms race with other countries. Who ever has bigger better ai will dominate and control the sector in the future. we get a super water downed version of the real deal. headless ai is a must
2026-06-06 19:02:01
10
ray cori :
Pied Piper - Silicon Valley 🫡
2026-06-07 05:44:15
9
Glassy Walls :
I will believe quantization is any good once I see a model that doesn’t choke on basic awareness.
2026-06-06 21:14:43
6
China :
It's still Google and I'll never use one of their products.
2026-06-08 15:33:36
4
⚡︎AıPrıest︎⚡︎ :
Pide Piper all over again..
2026-06-22 11:24:06
2
⃟ :
Still answers slowly on that Gemma 4 32B
2026-06-06 18:18:50
2
Lizard King :
Let me know once I can run deepseek V3 on a pair of 4090s
2026-06-09 17:41:48
2
LukePrivvs :
yes but that's compressing a 32 bit model. 32bit models are massive anyways and the standard is 4 to 8 bit anyways. 16-32bit models are never the standard as you have marginal gains
2026-06-07 10:37:27
0
mGGle :
Hopefully we can afford ram soon then
2026-06-08 13:24:42
1
Jesus Is King of Kings :
When you say shrink, do you mean like rewrite it? Because you can't shrink data like that. It just can't be done.
2026-06-09 01:31:24
0
To see more videos from user @whitewhoadie, please go to the Tikwm
homepage.
