ProTekk

First try

Qwen 3.7 27B/35B MoE is not going to happen. The window for that has passed.

@TMobile 5G is faster than my @verizonfios service. Sad.

‘Wi-Fi’ is not an acronym. It doesn’t stand for anything. Also! Since it’s a trademarked term, it should be written as ‘Wi-Fi’ (not ‘wifi’ or ‘WiFi’).

@no_stp_on_snek What did you pull it off at? What wood, rub, etc? Might need to try it out over my primes.

Awe yeah. Wagyu brisket is now. So jiggly.

@LottoLabs Need to submit a benchmark but I'm almost there. 65 pretty frequently with @spiritbuun's fork.

The biggest dark horse in all of ai right now Is qwen 27b on a 3090 going 70+ TPS Literally 6 year old tech that can think

llama.cpp now has an official website: llama.app Our goal is to make local AI accessible to everyone, and improving the user experience is a big part of that. On the new landing page you’ll find a single-line cross-platform installer. The installation provides a single unified `llama` entrypoint which you can use to run/serve models and interface with 3rd-party agentic applications. While oriented towards simplified user experience, the new `llama` application also provides all the advanced functionality of the existing llama.cpp tooling with which experienced users are already familiar. Also note that all GGUF models that you might have already downloaded with llama.cpp in the past will be automatically available to use without downloading again (they are stored in the common HF cache on your machine). We have many improvements in the pipeline both at the UX and at the engine level and we plan to iteratively ship new things over the coming months. One of the main focuses will be seamless integration with local-friendly 3rd-party agents (such as Pi). In the meantime, we’ll continue to listen for feedback from the community and adjust accordingly, so keep letting us know what you think and need.

The MAGAs have spent 11 years spreading hate around the internet like goddamn fertilizer but then clutch their pearls when someone tells them to shove it up their ass

Perfect.

@LottoLabs Problem is, at least in my experience FB marketplace sucks, negotiating a price, agreeing on a day/time then they tell you they sold it to someone else just before you leave to go meet up.

The deals are still out there

Uploaded by Qwen3.6 27B jinja template. Since 3.5 I've been slowly accumulating various updates from other's templates + adding in fixes of my own as I go. Decided I should probably share it. huggingface.co/spiritbuun/buu…

@LottoLabs First reasonably priced 3090/4090, it's mine.

Localmaxxing marketplace coming soon List your hardware for sale Include hardware runs on the ad, On-site messaging, nice link cards New users can easily see what type of models they can run if they buy or build a similar set up

Facebook marketplace is really the worst. Two sellers, one sold the card out from under me as I was prepping someone to go meet them, the other hasn't responded in days.

Have to shout out @spiritbuun and his llama.cpp fork. Qwen3.6 27B UD-Q5_K_XL + 262k ctx + turbo4 running on a single RTX 3090 headless is a great local setup and has been working perfectly. 180k ctx in, real world coding/app building and have not seen degradation.

I really can't today. prompt eval time = 5010.13 ms / 2099 tokens ( 2.39 ms per token, 418.95 tokens per second) eval time = 15449.23 ms / 126 tokens ( 122.61 ms per token, 8.16 tokens per second) total time = 20459.36 ms / 2225 tokens

This is all being done on a single RTX 3090 with a Q4_K_M quant and turboquant KV cache. Don't buy the bs from people saying you need SOTA, datacenters, f16-only to do "real" work. Local LLM/agents are very real and very capable. CC: @LottoLabs

Figured out how to really use Hermes and Qwen3.6 to do some real stuff. RE of Cobb AccessPort and flashing tunes to the ECU happening. Found private keys, dumping firmware and binaries, crazy stuff...

Alright, Qwen3.6 can do some stuff. Created this basic datalog viewer for my car in a couple of minutes.

@LottoLabs I think it's worth a shot. Local facebook marketplace post, just waiting for an answer on the when. If it works, need to upgrade my PSU for sure, 750 watts won't cut it.

@ProTekkFZS The fact it’s from somebody non-technical probably makes it a buy in my opinion. That’s the sweet spot for picking up those deals.

@LottoLabs Got a lead on a potentially bad vbios 3090 for $250. Worth the risk to potentially have 2x3090? I can do reflows and reflash. Only need it to work headless.