Het 👽

My X feed is full of guys fine tuning/ running models with what not optimisations on GPUs, TPUs, DGX locally, running agent swarms I also want to do these things, but fhuuuu😮‍💨 one day, one day

@soham901x @Het1501 Bro like bulky build 🤔

@het_bhalani @Het1501 My keyboard creates moderate noise, and I like it. When I was at the store buying this one, we saw a silent wireless keyboard from Portronics, and it was damn good. But I prefer the sound and bulky build 😆

A kid was curious, so I gave him my PC to try Minecraft Java Edition for the first time. He probably doesn't realize it yet, but sitting in front of a gaming PC and playing Minecraft like that was once a years-long dream of mine while watching others play on yt

@soham901x @Het1501 I’m using cosmic byte with blue switches, blue are not worth it, make more noise then expected, my frnd has kreo swarm with purple switches-it’s sounds like a butter and it feels like ahhhohohhoohhh🤤

@het_bhalani @Het1501 BTW, I've tried many keyboards so far. First was a budget zebronics keyboard, then a zebronics combo, then a logitech combo, which was actually good, and finally this mechanical keyboard with red switches

@soham901x —

@het_bhalani Yupwhynotwhynot

don't judge me guys 😆, yarn at home:

@soham901x @Het1501 Most gov. Keyboard u can think of

@het_bhalani @Het1501 youtu.be/ImU4RUutnlg watched this when I was 16-17yo, and it gave me urge to get it

@0xSero Next gen nvidia!

21 petabytes of memory bandwidth.

Me vs the guy I’m not worried about

@Het1501 @soham901x tvs is said to be one of the finest sounding keyboards...

@soham901x Tvs spotted 🫣

@soham901x still a dream

@soham901x @NILAY1556 🤫🤫

@het_bhalani @NILAY1556 Yup I agree, we aren't employed like others 👀

meet my friend

@soham901x 🤫

@het_bhalani BuzzBot

@NILAY1556 U suggested in the morning so i thought let’s give it a chance, baki to etla paisa kyathi kadhva?? We r not employed like other ppl! @soham901x 👀

@het_bhalani With api billing

@NILAY1556 No no, it’s default there, actually using nim via proxy

@soham901x @NILAY1556 Bro works in US shift, just to set in

@NILAY1556 🙂‍↔️🙂‍↔️, it is good morning, NILAY

Purple gradient op!! Noice , you gained more unusablity

@NILAY1556 @Hiteshdotcom bhagwan sanket aape 6 bhai👽

Well played @Hiteshdotcom Know your audience, currently placement seasons are running and ... youtu.be/cQYvpikHT8U

Anyone with 8GB or 12GB VRAM setups needs to understand that "-ncmoe" is the key flag to boost performance on llama.cpp Here are my results for Qwen3.6 35B A3B, with 64k q8_0 context on a 8GB RTX 3070Ti: ⚪️ no flag → 8.7 tok/s RAM: 13.6GB & VRAM: 7.8GB 🔴 -ncmoe 35 → 27.5 tok/s RAM: 12.1GB & VRAM: 4.3GB 🟢 -ncmoe 30 → 32.5 tok/s RAM: 12GB & VRAM: 5.6GB 🔵 -ncmoe 25 → 40.9 tok/s RAM: 12GB & VRAM: 6.9GB Please note the ram and vram usage you see are total usage of a windows pc, with the model running. My friend's setup: 8GB VRAM and 16GB RAM. You can boost performance by switching to Linux, just something to keep in mind. Basically, this flag keeps the MoE experts in the first X layers on your CPU + RAM, instead of eating all your VRAM straight away. This is a smart hybrid offload way that lets you run bigger models without OOM while keeping the rest on your GPU for speed. As we can see on the data, there's a sweet spot. When we lower it from 35 to 25, speed bumps +50% because there are more layers on your GPU (look at the VRAM usage). The key here is to play around with the number and fit as much as possible on your VRAM, goal is to have 1GB/800MB headroom to avoid stress. ↓ server flags below

@digitalix Spark like literally

Which DGX Spark is running Gemma 4?