Mastodawn

RE: https://mastodon.social/@clifmo/116456465453931672

Just wrapped a fully local LLM stack on Fedora + AMD RX 7900 XTX (24GB VRAM).

🔹 Engine: llama.cpp w/ Vulkan/RADV backend
🔹 Model: Qwen3.6-35B-A3B (MoE, Q4_K_M, ~21.4GB)
🔹 Config: -ngl 99 --ctx-size 16384 --batch-size 512
🔹 Perf: ~137 tok/s gen, ~344 tok/s prompt, rock-solid at 16k context
🔹 All VRAM. Zero CPU offload.

Powering AnythingLLM for private RAG pipelines. 🐧⚡

#LocalLLM #Fedora #AMDGPU #Vulkan #OpenSourceAI #RAG #LLM

Clifton (klif+THUN) he/him Apr 23

Installed #Fedora Workstation today. First time using it. Good bye #Ubuntu. Mainly for first class #Vulkan and #Pipewire support. Going to try out #Qwen 3.6. #Gemma4 did great captioning my #Photoprism archive given it ran on a base model M4 Mac Mini. Time to see what the big rig can do.

Clifton (klif+THUN) he/him Jan 22

I bet none of you received a Kazakhstan gift set of chocolates & Ablai Khan (Likely) journal

Clifton (klif+THUN) he/him Jan 22

Paws and Pause - from Nov 20, 2021

A fluffy tabby kitten lounges on a carpeted cat tree, peeking around a wooden post with wide, curious eyes. One front paw dangles lazily over the edge as she relaxes.

Clifton (klif+THUN) he/him Jan 18

Anti-Futures: Resisting the Hype Cycle - In a future that feels remarkably anti-future, we need digital tools that really work for real people. Good can befriend imperfection. Let's show up for each other in 2026.

https://ethicaldigitallabs.com/essays#letter

The Library | Ethical Digital Labs

Clifton (klif+THUN) he/him Jan 2