Stories by heggenhougen

Show HN: 1k desktop beats vendor sparse library 474× on Mistral-7B

LLM FFN benchmarks on a 4‑core HP All‑in‑One

Benchmarking rolvsparse on DeepSeek-R1 and Llama 4 – up to 82x vs. cuBLAS

Show HN: ROLV – 20x faster MoE FFN inference on Llama 4 Maverick vs. cuBLAS

ROLV: 50.5× speedup and 91.4% energy savings Qwen2.5-72B MoE expert FFN (1xB200)

100% secure way to test proprietary sparse AI kernel (ROLV) on any hardware

ROLV – Beats Vendor Kernels, Cross‑Platform

Rolv.ai: Universal sparse compute primitive, back end-agnostic reproducibility

The ROLV CPU Breakthrough