Home | Notifications | New Note | Local | Federated | Search | Logout

Note Detail

lain@lain@fediffusion.art (2026-09-14 12:52:09)
Post from an instance 100% vibecoded in ~2 days with qwen 3.8 flash next, local inference on a strix halo system.

RT: https://freefed-live.tunnel.eosrift.com/posts/01M2F0B3MJNCV6K5KT9PDFW44Z
---Reply---
underproofed ganbold@ganbold (2026-09-14 13:17:49)
how do you use qwen 3.8 flash next, lain? like is it for all-in-one model for you? or you just delegate it only for implementation task?

the amount of params itself interest me, 125B-A6B. I tried 26B-A4B and 35B-A3B previously since my ram is not enough. but kinda looking forward to try qwen 3.8 flash next as well

---Replies---

lain@lain@fediffusion.art (2026-09-14 13:24:12)
@ganbold yes, this one was all-in-one, i just gave it the task and it worked on it until it's done, with multiple compactions in between.