Francisco Abner

Electrical Engineer. CEO & Founder of YOSO-YAi.

Electrician turned EE. I do surgery on open-weights models and measure Energy Per Intelligence on production ARM hardware.

Fluor > Tesla > Meta > YOSO-YAi. The future of labor is compute.

Research Focus

  • Energy Per Intelligence (EPI) -- A metric that measures useful intelligence produced per joule of energy consumed
  • Model Surgery -- Expert pruning, mixed quantization, attention head removal -- evaluated by EPI on real ARM hardware
  • Embedded AI -- Running optimized models on Raspberry Pi clusters and custom PCBs

Open Research

Repo Description
energy-per-intelligence Defines the EPI metric. Baseline measurements on Pi cluster.
epi-meter Open-source 4-channel AC power measurement board (RP2350 + ATM90E26)
epi-bench Calculate EPI, plot Pareto frontiers, log power traces, run benchmarks
expert-pruning-epi MoE expert removal measured by EPI (Qwen3-30B-A3B on Pi 5)
mixed-quant-epi Per-layer quantization -- Pareto frontier of accuracy vs energy
attention-head-surgery-epi Attention head removal -- do remaining heads compensate?

Links

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support