Founding CTO / GPU Systems Engineer at Syntropic
We compress LLM KV cache. Measured: 8× at 3-bit with 0.43% quality delta (GPU-validated); 22× live end-to-end on a 5-tenant workload, GPU-verified this week...
We compress LLM KV cache. Measured: 8× at 3-bit with 0.43% quality delta (GPU-validated); 22× live end-to-end on a 5-tenant workload, GPU-verified this week...