Alex Dimakis(@AlexGDimakis):Introducing SkySynth: Enough with general-purpose systems that support many workloads and hardware configurations. SkySynth can write (on the fly!) an inference engine specific to the model/GPU and workloads you want to serve (eg Qwen3-4B 2.2x faster throughput compared to SGLang and vLLM) and a custom router that is 2x lower cost compared to generally optimized routers. Finally, a router that cares about your personal needs :p

2026.09.11 19:19

Introducing SkySynth: Enough with general-purpose systems that support many workloads and hardware configurations. SkySynth can write (on the fly!) an inference engine specific to the model/GPU and workloads you want to serve (eg Qwen3-4B 2.2x faster throughput compared to SGLang and vLLM) and a custom router that is 2x lower cost compared to generally optimized routers. Finally, a router that cares about your personal needs :p

显示更多

Shu Lynn Liu@shulynnliu

2026.09.11 18:11

Agents let us build systems for different workloads and requirements. But… can we trust what they build? We release 🌟SkySynth🌟: an engine for synthesizing high-performance, just-in-time (JIT) systems we can trust, by co-evolving formal proofs and tests alongside the code. Results: 💿 KV stores up to 2.3× faster than Redis and FASTER + formally verified stores with 2.9× Claude Code's pass rate 🚏 Model routers up to 48% cheaper than a general router ⚡ Specialized inference engine with 2.2× the throughput of vLLM/SGLang 🧵👇

显示更多