Juan Pendragon (@Dev_Creator8) header
Juan Pendragon (@Dev_Creator8) profile picture

Truth is not always pretty but it would make you free. La verdad no siempre es bonita pero te hará libre.

Joined April 2021
Photo by Juan Pendragon (@Dev_Creator8)Photo by Juan Pendragon (@Dev_Creator8)Photo by Juan Pendragon (@Dev_Creator8)Photo by Juan Pendragon (@Dev_Creator8)Photo by Juan Pendragon (@Dev_Creator8)Photo by Juan Pendragon (@Dev_Creator8)Photo by Juan Pendragon (@Dev_Creator8)Photo by Juan Pendragon (@Dev_Creator8)Photo by Juan Pendragon (@Dev_Creator8)Photo by Juan Pendragon (@Dev_Creator8)
DeepSeek flash the best, what Claude refuse to do DeepSeek did it in a few minutes. I have my tv clean now.
65
Stupid claude doesn't want to help me root my tv, any advice here?
54
I tried this on my RX 6700 XT 12GB, using the same benchkit. Only passed 6 out of 92 and couldn't finish the whole suite due to segnentation fault. This was with even lower context: 65K and only 25 t/s on average.
Qwen3.8-27B on a single RTX 3060. after weeks of testing quants and configs, this is where I'm stopping -- these two are my final builds for this card. 158K ctx at 22 tok/s, or 82K at 38 tok/s. (config & weights below) tested four IQ2 quants across seven benchmarks and GSQ-RCO IQ2_XS won, while being the second smallest file of the four. the catch: GSQ-RCO ships no MTP head. you borrow Unsloth's -- same base model, drops straight in -- and pay ~76K of context for 1.7x decode. prefill is ~460 tok/s either way, MTP only touches generation. which one is "worth it" depends entirely on what you run. short few-shot tasks never touch 64K, so the extra context is dead weight and you just want the speed. long single-session agent runs blow past 82K and no decode rate saves you once you're out of context. that's why both configs exist. everything on low reasoning effort. at 2-bit the higher settings spiral instead of terminating. 🐢 context (158K): llama-server -m Qwen3.8-27B-GSQ-RCO-IQ2_XS.gguf -ngl 99 -c 158000 -fa on --jinja -np 1 --cache-type-k q4_0 --cache-type-v q4_0 --temp 1.0 --top-p 0.95 --top-k 20 --min-p 0.0 --presence-penalty 0.0 --repeat-penalty 1.0 --reasoning on --chat-template-kwargs '{"reasoning_effort":"low"}' ⚡ MTP (82K): llama-server -m Qwen3.8-27B-GSQ-RCO-IQ2_XS.gguf -md mtp-Qwen3.8-27B-Q4_0.gguf -ngl 99 -c 82000 -fa on --jinja -np 1 --spec-type draft-mtp --cache-type-k q4_0 --cache-type-v q4_0 --temp 1.0 --top-p 0.95 --top-k 20 --min-p 0.0 --presence-penalty 0.0 --repeat-penalty 1.0 --reasoning on --chat-template-kwargs '{"reasoning_effort":"low"}'
1
73
Ok, so after a lot of effort trying to make this setup work I'm ready to call bulshit on this. 1. The commit hash in the screenshot doesn't exist on that repo ("atomicmilkshake/llama-cpp-turboquant"), or any fork I could find. I had better results with standard llamacpp 🧵
Hello, yes of course. This is what I'm running currently. The screenshots are from the 24th, and they're a bit fuzzy from cropping, sorry. It includes info about a few bugs I had to figure out but nothing too bad.
2
141
With regular llama cpp I could achieve this: llama.cpp: commit 5f55650 Model: Qwen3.6-35B-A3B-UD-Q4_K_M Context: 65k Speculative decoding: enabled Performance: ~32–36 tok/s Prompt processing: ~700 tok/s
1
23
./build/bin/llama-server \ -m ~/modelos_llm/Qwen3.6-35B-A3B-UD-Q4_K_M.gguf \ -t 10 \ -ngl 99 \ --n-cpu-moe 30 \ -c 65536 \ -fa on \ -np 1 \ --cache-type-k q8_0 \ --cache-type-v q8_0 \ -b 2048 \ -ub 1024 \ --spec-type draft-mtp \ --spec-draft-n-max 2 \ --host 0.0.0.0 \ --port 8080
23
I'm starting to work with local llms, I have a RX 6700XT VRAM 12GB and 32GB of RAM. I'm using llama.cpp on linux and right now I testing Qwen3.6-35B-A3B-UD-IQ4_XS.gguf, these are my results so far: Tk/s: 7.2 Used memory: 11577 average Temperature: 57°C
1
105
Flags used: Qwen3.6-35B-A3B-UD-IQ4_XS.gguf \ -ngl 24 \ -c 32768 \ --cache-type-k q4_0 \ --cache-type-v q4_0 \ --flash-attn on \ -t 6 \ --host 0.0.0.0 --port 8080 Is this the max I can expect or is there a way to improve it even further?
75
Los fanáticoa dirán que s IA.
Este vídeo dura casi 8 minutos y demuestra cómo Abelardo ha acomodado sus posturas para obtener más votos. Es un estafador que lo único que quiere es llegar al poder como sea. Lástima que en lo único que no haya sido capaz de cambiar es en no usar mocasines sin medias. Que boleta.
1
51
One of the hardest photos I've taken.
one of the hardest photos i have taken
92
Ok, first attempt to make the water. 12 Vdc adapter. Thin food plastic bags. 15 seconds contact (close circuit) Nothing extraordinary so far. Maybe a I didn't pull out the bag fast enough but I don't see how I can do it faster to be honest.
1
7
1,385
I don't know if this is true but I like it.
"Turning the other cheek" isn't about submission. It is an act of defiance. Modern society interprets Christ's command as a call to be a doormat, to be passive, weak, and defenseless. In 1st Century Judea, the left hand was considered unclean. You would never use it to strike someone. Therefore, to slap someone on the right cheek (as Jesus specifies), you had to use the back of your right hand. A backhand slap was an act of insult. It was how a superior hit an inferior. It said, "You are beneath me." By turning your head and offering the left cheek, you physically prevent them from backhanding you again. You force the aggressor to use their open palm or a fist. Why does this matter? Because a fist is how you hit an equal. Jesus wasn't telling his followers to cower. He was telling them to stand their ground, look their oppressor in the eye, and say without words: "I am not your slave. I am a man. If you are going to hit me, treat me like one." It is not an act of cowardice. It is an assertion of dignity that requires terrifying courage.
56
Timeline cleaning.
The best video on the internet today
28
Alien ships don’t use 1,000,000,000 watts of power to bend spacetime. They manipulate the hologram with 1,000,000,000x less power than our physics says is needed.
35
15
8
225
14,155
I know I'm not a native English speaker, but is it that bad that I can't understand whatever is written there?
Rockstar confirms that GTA 6 is there were once they are did. They became the first ever to have the first as in the video game of that when of PS5 began. Congratulations Rockstar!
83