Rysana
42 posts
Rysana
@Rysana
We create the world's fastest and most powerful AI.
Joined June 2023
- Rysana reposted
Introducing Inversion, our family of structured LLMs. Our first generation models excel in structured tasks, offering unmatched speed, latency, reliability, and efficiency, with the most comprehensive typed JSON output support available anywhere. rysana.com/inversion
00:00 - Since this post, we've breached 100,000 t/s (per user) on optimized workloads and are targeting a peak of >1,000,000 t/s in production by EOY.Over 9000! tokens per second. Ultra-fast, always-valid typed JSON with LLMs. rysana.com/log/over-9000-…
- Fastest LLM API in the world, end-to-end function calling in the blink of an eye
- Not too long ago, electricity was mostly confined to lightning here and there. Last century, it became available and malleable for humans, first sparsely and now with great abundance all over. Intelligence is this century's electricity. Let's put it everywhere, in everything.
- Rysana repostedinversion-sm is now, on average, 2.39x faster, 14.4% smarter, and 2.11x cheaper than when we announced the model, and completes the average request in well under 200ms. that's a +475% boost in intelligence flux since late March, and >1000% since the first checkpoint.
- Rysana repostedin the past month: ~4.5× faster inversion compiler ~100× faster sampling ~2× faster runtime overhead ~12% faster overall speed, same models ~6× faster query parser ~95% less internal networking
- Applied general intelligence flux (IQ*Hz/$) will increase by over 1,000,000,000x by the end of this decade.
- Rysana repostedToday, we've optimized the Inversion compiler's runtime overhead again, pushing down below ~9 microseconds on even the most complex structures. This work builds out the support for up to 100,000 tokens/second inference with perfect structured output.
