News for hackers. Stuff that matters.
Re:Step 5 Preview, a 1M-context MoE from StepFun, shows up o... by UncleOxidant on Thursday October 08, 2026 @10:23PM (#50015213)
Seems like it was an under-noticed model back when it came out because there were so many new Qwen models coming out around the same time. I tried it some on my Strix Halo box, but it was right on the edge of fitting.
Re:Step 5 Preview, a 1M-context MoE from StepFun, shows up o... by theturtletalks on Friday October 09, 2026 @12:59AM (#50016168)
3.5 Flash was decent, but I was using it on OpenRouter. What model are you running on your Strix Halo nowadays?
Re:Step 5 Preview, a 1M-context MoE from StepFun, shows up o... by Iolaum on Friday October 09, 2026 @01:49AM (#50016482)
Qwen-3.8-flash-next @Q4
Re:Step 5 Preview, a 1M-context MoE from StepFun, shows up o... by throwa356262 on Friday October 09, 2026 @09:00AM (#50019856)
Not sure why this is downvoted.

It is fully possible to run the 180B Owen model on strix halo using strata and even get reasonable speed out of it.

https://github.com/Niko1221/Strata

Re:Step 5 Preview, a 1M-context MoE from StepFun, shows up o... by Iolaum on Friday October 09, 2026 @09:26AM (#50020225)
Indeed - Even with llama.cpp you can store n-grams to disk and use it. I m using unsloth's llama fork atm but a new llama release is out that may be better - hadn't had time to test it yet.
Re:Step 5 Preview, a 1M-context MoE from StepFun, shows up o... by UncleOxidant on Friday October 09, 2026 @10:47AM (#50021281)
Yes, this on Halogen. I can get ~45tok/sec which is quite usable.