David Zachariah

Computer & Electronics Enthusiast | Full Stack Developer in JS | Into Information Security | Wanna make the world a better place

[David Zachariah @davezaxh]

i've started to learn about quantising llms, would be putting my findings in this thread. if you guys know any relevant resources, feel free to share

[David Zachariah @davezaxh]

@tonychenAI @KleePay sent you a dm!
i shipped an on device (desktop + mobile) voice memo proactive ledger that gives you a log of your daily thoughts with zero data leaving your device

[David Zachariah @davezaxh]

@Prajwxl particularly useful if you're running a nas

[David Zachariah @davezaxh]

@its_airpods69 sent you a dm

[David Zachariah @davezaxh]

@Prajwxl wassup!

[David Zachariah @davezaxh]

@photon_hq built an AI agent for incident triage, a coding tutor that gives hints not answers, and a file scanner. working inside an applied AI lab right now. i know how the tech works so i can tell the story right
https://t.co/Nu6GzCeS6x

[David Zachariah @davezaxh]

@JunaidAckroyd @WorkWeave built an AI agent for incident triage. you describe whats wrong in plain English, it checks logs and configs, figures out what needs fixing and does it. human in the loop so you approve before anything runs.
Next.js · Python · CI/CD
more at https://t.co/Nu6GzCeS6x

[David Zachariah @davezaxh]

@wh0sumit @rightfitso check out my portfolio on: https://t.co/Nu6GzCeS6x

[David Zachariah @davezaxh]

@Hrushikeshhhh sent you a dm

[David Zachariah @davezaxh]

@IronRedSandHive @Prajwxl it's actually pretty simple, let me know if you need any help

[David Zachariah @davezaxh]

@Prajwxl build times must be crazy long

[David Zachariah @davezaxh]

@huggingface for my calibration texts I compiled a bunch of snippets of python and javascript from popular github repos such as flask, nextjs and fastapi which was around 134kb in size

[David Zachariah @davezaxh]

i've started to learn about quantising llms, would be putting my findings in this thread. if you guys know any relevant resources, feel free to share

[David Zachariah @davezaxh]

@huggingface this matrix is then used during quantization to prioritize important weights with higher precision, while less important ones are compressed further, preserving quality while shrinking the model

[David Zachariah @davezaxh]

@ayrusReT @huggingface Quality holds up way better than expected, didn’t see real degradation until Q3_K_M, mostly on longer multi step reasoning.