Interesting: i gave claude and Qwen 3.5 9B the same problem to solve do a small home assistant automation. Qwen was running on my 5.5 year old Laptop (8 core Ryzen, 32gig Ram no useable GPU(at least i think so)) claude was done in 20sec, my Laptop needed 240sec but the endresult was nearly identical and correct. Also the annotations were nearly identical. Okay it is slow but better use my green power here than using claude which site in some coal powered AWS etc datacenter
#ai #janai #qwen3.5
#qwen3
3 posts · Last used Aug 29
fazit nach ner woche #programmieren mit #opencode und #qwen3.6 27b über #llamacpp :
macht richtig spass! 😎
man muss das teil zwar eng an der hand nehmen und muss gucken dass man immer nur ganz kleine dinge in auftrag gibt, aber wenn man sich mal darauf eingestellt hat, dann ists tatsächlich - und immer noch bisschen unglaublicherweise - produktiv nutzbar
auf consumer hardware im eigenen #homelab - crazy
New update for the slides of my talk "Run LLMs Locally": WebGPU
Now models can run completely inside the browser using Transformer.js, Vulkan and WebGPU (slower than llama.cpp, but already usable).
https://codeberg.org/thbley/talks/raw/branch/main/Run_LLMs_Locally_2026_ThomasBley.pdf
#ai #llm #llamacpp #stablediffusion #gptoss #qwen3 #glm #localai #webgpu
You've seen all posts