Lisan al Gaib criticizes Wafer.ai's self-promotional claim of running GLM 5.2 on AMD hardware at 80% of B200 throughput

@scaling01 It works + they will prob raise off that

@xeophon unfortunately slop works

@scaling01 I saw the 200+ tps was at 10k input and 1.5k output, not sure it would hold up at long context as b300

@scaling01 They just need their CEO to have a based X account.

@scaling01 don't forget the toolcall error rate also being awful

@scaling01 I see you always posting against open source. Looks like you've a beef against chinese or people paying u well for poating this type of posts

@scaling01 i mean if it works you gotta respect the game

@0xSmite @scaling01 actually you can dislike it when people poison the commons even if it's profitable

@xeophon @scaling01 yes

@scaling01 @xeophon Slop is all you need for a billion

@scaling01 this trend of manufacturing drama for engagement is getting old

everyone just wants to be the main character of their own mess

@scaling01 Kinda insane marketing AI agents to inference engineers

@scaling01 Yea the PR is cringe. But GLM 5.2 on AMD at that price point is a real signal for production inference.