Viewing profile — unrvl22

its kinda crazy with literally no guardrails and a goal, the extremes these AI models can actually go to.

a good take.

Something like this is nice, where instead of having 1 model with X active experts, you have 10 different models, all small and dense, trained on specific information. and loaded o…

inference is only memory bandwidth limited when targeting higher tps / high single stream tps. the weights only need to be moved across once per forward pass, when you batch say 10…

that 213 wasn't achieved when saturated though. was probably more like 30 tps per stream when doing 2.6k tps.

MI355X can perform FP6 operations with the same speed as their FP4 (unique to AMD) - people should be making MXFP6 quants which would be pretty much lossless, and much closer to FP…

look at benchmarks, use the model yourself. Im usually first to call BS on every chinese model that says they are as good as Opus. this is finally the first one that actually is. I…

the 2 I mentioned both have a fairly large following, who run benchmarks and absolutely will spot issues.

I cancelled my claude sub after realizing I can burn 300m tokens a day of this quality, for $50 a month.

Why aren't more people talking about this? It's literally Opus 4.7 quality stupid prices. I know providers who are offering this at unlimited tokens for $50 a month. Some are even …

The municipality of Rio de Janeiro (via its IT company IplanRIO) released Rio-3.5-Open-397B, presented as a homegrown Qwen3.5 fine-tune that beats comparable open models on benchma…

AI slop. looks basic as hell.

why is deepseek v4 pro a lot lower than flash? where is mimo 2.5?

I had no idea those dan and the team were aussies! damn nice, we dont really seem to shine in tech on the world stage.

love how it loads instantly and feels smooth. imo useless but still cool