@github.awesome: Colibrì runs a 744-billion-parameter model on a laptop with 25GB of RAM, in pure C with zero dependencies. A model that size normally wants hundreds of gigabytes of memory or a rack of GPUs you'll never own. The author's blunt that it's slow, roughly a tenth of a token per second on his humble box, but it's a frontier model answering correctly on hardware cheaper than an H100's fan. #github #opensource
a 500 token response would take almost an hour and a half.
2026-07-11 01:44:05
112
Sebastian Rolando :
at the end of the response "42"
2026-07-12 21:16:00
0
the man of the hour :
Horribly impractical but getting GLM 5.2 to run on consumer hardware is no joke.
2026-07-11 04:24:48
36
Music :
currently the best thing on 96-128 ddr5 ram with rtx 5070ti laptop is deepseek v4 flash from antirez/ds4
2026-07-11 20:29:14
2
joe :
not as good as just using gwen3.6-27b
2026-07-11 07:38:32
11
mead-bee :
Moe is the future. This is 40 which is solid for nearly anything..Even the 3b moe's are impressive
2026-07-12 02:52:51
1
Planck’sVariable :
Can you put it on rtx4090? Maybe if it was a bit smaller. Any suggestions for the best next thing?
2026-07-11 04:37:51
1
Flo :
too slow for everyday use but very impressive engineering! i hope this evolves into something great!
2026-07-11 11:54:24
5
Dustin :
it's so glacierally slow I had to delete it.
2026-07-11 01:48:57
21
aris🫰🤌🤌 :
I'm so impressed it's made in C no one wants to code in C this days I'm sure they are super smart prof as some university
2026-07-11 01:33:51
2
Shabby :
now let's see that inference speed 😂
2026-07-11 10:54:57
0
steven :
0.001 tokens / second
2026-07-12 10:56:41
0
Kharbasha :
2 min for 2 line code 😭
2026-07-11 18:23:45
1
Bolvarsdad :
holy vibe code
2026-07-12 00:59:16
0
Chokmah777 :
i almost thought someone beat me to it
2026-07-12 00:14:01
0
ɔiɿɘᎸꙅoƚɒɿƚꙄ ɔiɿƎ :
I never had enough patience to wait for the answer.
2026-07-11 23:40:23
0
gatesofshawn :
the question is whether the more hard or you throw at it the faster it gets so could I get one token per second or five even I just adding some gpus or do I need to add like a giant GPU before it even makes a difference?
2026-07-11 19:07:17
0
RandomBro :
This is the future!!!
2026-07-11 01:19:02
1
tabbytabby224 :
would be useful for remote areas in poor countries without internet connections. you could ask in depth medical questions and get an answer in a few hours.
2026-07-11 07:56:39
5
h :
slop
2026-07-11 20:50:47
0
To see more videos from user @github.awesome, please go to the Tikwm
homepage.