Research & engineering

Notes from the lab.

The techniques inside FlowServe — quantization, speculative decoding, anchored fine-tuning, hardware-agnostic scheduling — written up in the open. Papers, engineering deep-dives, and reproducible benchmarks.

Featured paperPreprint · forthcoming

TurboQuant: 1–4 bit KV-cache quantization for memory-bound LLM inference

Aggressive, runtime-toggleable KV-cache compression up to 16× that treats memory bandwidth as the first-class bottleneck of quantized serving — trading footprint for negligible accuracy loss.

FoundationFlow Research·2026·14 min

All writing

Get new research in your inbox.

Preprints, benchmarks, and engineering notes as we publish them. No marketing.