Owais

i do stuff

[Owais @muffBozo]

it would always anger me whenever i would get stuck on a topic while watching youtube lecture or during my JEE days the LMS lectures of my coaching

[Owais @muffBozo]

@adityadotdev hey idont think you have dms open for everyone, couldnt see it on your profile

[Owais @muffBozo]

@ElevenLabs @ElevenLabs you should checkout gUrrT we do everything you do but in way less compute and way more deeply
https://t.co/a5FHyc4TYl

[Owais @muffBozo]

@neural_avb you should checkout gUrrT

YouTube ask is cool but what if we could do everything ask does on just 4gigs of vram
that is what gUrrT solves

https://t.co/a5FHyc4TYl

[Owais @muffBozo]

For everyone interested in contributing, gUrrT is completely open source.
Whether it is a bug fix, a feature idea, or just poking around the codebase, all of it is welcome.
GitHub repo below.
https://t.co/a5FHyc4TYl

[Owais @muffBozo]

Releasing gUrrT v2!! a local, open source Conversational Video Intelligence for Q&A over all of your video lectures.

No GPU required. No subscriptions. Your video never leaves your machine.
#OpenSource #LocalAI #VLM #LLM #VideoAI #Python

https://t.co/dvuNOwXqyL

[Owais @muffBozo]

gUrrT's job is to find the right frames from inside a video the ones where content actually changed. Do it fast, do it clean.
v1 was blind on every video over 23 minutes.
v2 is not.
Same videos. Same task. Very different results.

v2 Releasing SOON!!!

[Owais @muffBozo]

v2 finds it.
4x faster. 100x cleaner.
Full breakdown on Medium
https://t.co/roKLuOLa56

[Owais @muffBozo]

gUrrT v1 had two ways to extract frames from a video lecture.
Both failed.
Here is what that looked like 🧵

[Owais @muffBozo]

A 1 hour lecture at 30fps is 108,000 frames.
Under 1% of them actually contain new information.
v1 could not find that 1%.

[Owais @muffBozo]

@ElevenLabs @PhysicswallahAP @physics__wallah Checkout gUrrT -> it is opensource local first your data stays on your servers
https://t.co/dvuNOwXqyL

[Owais @muffBozo]

@ElevenLabs @PhysicswallahAP @physics__wallah you are doing worse for your team by sending information about your vidoes to eleven labs servers who are surely going to sell their data, that is your proprietary data being sold

[Owais @muffBozo]

llama cpp has played a key role in shaping gUrrT v2
do checkout

[fareha @as_farrr]

I just published How switching to llama.cpp made captioning 2-4X faster https://t.co/xS4MD4sm3D

[Owais @muffBozo]

@as_farrr onwards and upwards.