Luigi Maselli

@lu_zero_ @antirez it mostly depends on the LLM model and the quality of data in the context provided by tools/pi-extensions youtube.com/watch?v=A_fvWh… For simple tool usage, you don't need a 10K context to explain how to use tools

@grigi0 @antirez but it is as good?

Look at this. Also opencode uses freaking 11k tokens of system prompt. Even at decent pre-fill of ~130 t/s it means waiting 84 seconds to start a session. What's the point? :D The pi agent is a lot saner here. Moreover, one could say, let's cache on disk very long common KV cache chunks, no? Hash it with all the parameters and put a sensible TTL if not used. But also: only cache it if you see it repeated N times across different sessions.

Artix linux > #archlinux still strong on distrowatch youtube.com/watch?v=NGla_1…

@mudler_it @lu_zero_ huggingface.co/mudler/Qwen3.6…

Tested Qwen3.6-35B-A3B-APEX-I-Compact, almost as good as Qwen3.6-35B-A3B-UD-Q4_K_M but it takes 4Gb less. //cc @mudler_it @lu_zero_

@UnslothAI @Alibaba_Qwen @opencode Just beware chat template for thinking reddit.com/r/LocalLLaMA/c…

@UnslothAI Yes Qwen3.6 35B A3B UD Q4 K M is the new champ also in my local bench! Amazing! 🎊 @Alibaba_Qwen @opencode

Qwen3.6 35b a3b perform worse than Qwen3.5 in my benchmark, but it should be something related to the quantization or tool calling //cc @UnslothAI gemma4 26b a4b still better for me #localai #llm

There isn't aclear winner but qwen is faster grigio.org/are-local-llms…

Are Local LLMs good enough for Vibe Coding? Gemma4-26B-A4B vs Qwen3.5-35B-A3B

nothink is the best tradeoff between accuracy and speed

Gemma 4 26B A4B is currently the best tradeoff in agentic benchmark for GPU poor people, specially the nothink version

Create your personal benchmark with llm-eval-simple, specially for local llm github.com/grigio/llm-eva…

Create your personal LLM benchmark with Opencode Benchmark Dashboard github.com/grigio/opencod…

Currently qwen-3.6-plus-free is the best model you can run on #opencode for quality and speed.

@lu_zero_ Purtroppo ci sono progetti che uso che dipendono strettamente da systemd github.com/cockpit-projec… Eccezioni a parte, #dinit funziona bene sia per processi di sistema che per sessione utente

@grigi0 io sono ancora felice con openrc e questo mi ricorda che c'e` la nostra implementazione rust di mdev da aggiornare e provare un po' di piu`.

Un Linux senza #systemd è possibile #dinit #artixlinux

LINUX è stato compromesso? systemd e la verifica dell'età grigio.org/linux-e-stato-…

artix / labwc / dinit youtube.com/watch?v=tDuR5L…

Is possible a modern linux without #systemd ? #ageverification