@github.awesome: Every "small" language model that does function calling still wants gigabytes of RAM and a real GPU. Needle by Cactus Compute is 26 million parameters, small enough to run on a microcontroller, a Raspberry Pi Zero, or any embedded device you've got lying around. Built specifically for tool calling, it picks the right function and fills in the arguments without phoning home to OpenAI. Smart home gear, IoT, wearables, all the places where shipping inference to the cloud is just not an option. #github #opensource
Ok but you skipped over the most important question which is how well does it work? Sure it can call tools but does it call the right tools and how often does it make mistakes? Can it reason at all? What is the size of the context window etc.
2026-05-13 11:59:00
13
me36776433 :
are we calling it jigabytes nowadays
2026-05-17 14:26:20
4
Greg :
Ah yes, the old jigabyte problem
2026-05-13 19:45:33
5
سيف | Gurt :
i love you man
2026-05-15 21:27:23
0
whatever consciousness is :
bro ur vidoes are awesome thank you
2026-06-19 16:49:34
0
Ali Malik :
I read the needle repository. none of this is there. even on their website.
2026-05-14 20:35:25
1
Elender Góis Gallas :
I don't understand the problem with the word gigabytes. How do you pronounce it then? I know for sure it is nor a hard g.
2026-06-04 02:52:52
0
Steven Stark :
Gemma 4 E2b / e4b
2026-06-26 17:25:15
0
Xaolin Showdown3 :
hmmmm 🥰
2026-05-14 13:04:27
0
Korpip :
Man all these AI tools and not a single showcase of it doing anything.
2026-05-15 17:03:19
4
Maxin :
👍👍👍
2026-05-27 00:27:50
0
To see more videos from user @github.awesome, please go to the Tikwm
homepage.