SutroYaro¶

Research workspace for the Sutro Group, a study group exploring energy-efficient AI training. Weekly meetings at South Park Commons, San Francisco.

What this site is for¶

SutroYaro is the lab's memory and dispatcher — not where the cutting-edge research happens, but where you go to find out what's been tried, who's working on what right now, and how to spin up new builds against the org's other repos. Five things it does:

1. Lab memory — DISCOVERIES.md is the curated "what's proven" record.
2. Agent-teams dispatcher — the hinton-problems and schmidhuber-problems catalogues (111 stubs combined) were built from SutroYaro sessions; the reusable machinery lives here.
3. Cross-repo index — see Related Repos for the full 8-repo map.
4. Public face — this site, shown at the Monday meetings.
5. Telegram + Google Docs + GitHub sync — see the Sync Runbook.

The active research front itself has migrated to cybertronai/ByteDMD/experiments/grid. Benchmark problems live in their own per-purpose repos. Cost metrics are defined in ByteDMD and simplified-dally-model. SutroYaro is the integration layer that keeps everything visible and connected.

The Challenges¶

The group runs several energy-efficient-learning challenges in parallel, each in its own repo:

See the Challenges index for current records, who is active, open questions, and where to start on each.

Where the Active Research Lives¶

This repo is the lab notebook for the group's Phase 1 and Phase 2 work. The live research front has migrated to cybertronai/ByteDMD/experiments/grid — Yaroslav's collection of self-contained experiments built around the new ByteDMD metric. If you want the current edge of the work, start there. SutroYaro stays the home for accumulated findings, the autonomous-research lab, the public site, and the sparse parity challenge.

What We Found¶

36 experiments across three phases, plus GPU energy validation. The full ranked results are in the Practitioner's Field Guide.

Phase 1 (16 experiments): Started with a broken SGD baseline (LR=0.5, stuck at 54%). Fixed hyperparameters to solve it in 0.12s. Optimized data movement within the SGD framework, hitting a ceiling because one tensor (W1) dominates 75% of all float reads. Pivoted to new algorithms.

Phase 2 (17 experiments): Tested algebraic, information-theoretic, local learning, hardware-aware, and alternative approaches in parallel:

DMD = Data Movement Distance (Ding et al.), measured via TrackedArray. Lower is better.

All four local learning rules (Hebbian, Predictive Coding, Equilibrium Propagation, Target Propagation) failed at chance level. Parity requires k-th order interaction detection, which local statistics cannot provide.

GPU measurement: Ran methods on NVIDIA L4 via Modal Labs using PyTorch CUDA (5 runs). GPU is 4-790x slower than CPU at this problem size. Sparse parity tensors are too small for CUDA to help. See GPU vs CPU findings.

Quick Start¶

git clone https://github.com/cybertronai/SutroYaro.git
cd SutroYaro

# Option A: Nix (recommended)
nix develop

# Option B: pip fallback
export PYTHONPATH=$PWD/src:$PYTHONPATH && pip install numpy

# Verify all 14 experiments across 3 challenges in <1 second
python3 bin/reproduce-all

# Run sparse parity with GF(2) (509 microseconds)
python3 src/harness.py --method gf2 --n_bits 20 --k_sparse 3

# Run sparse sum (new challenge)
python3 src/harness.py --challenge sparse-sum --method sgd

# Measure real GPU energy via Modal Labs
pip install modal && modal token set
modal run bin/gpu_energy.py

# Run autonomous agent loop
bin/run-agent --tool claude --max 10

Full setup instructions in CONTRIBUTING.md.

Where to Find Things¶

Quick Links¶

Companion baseline catalogs (shipped May 2026)¶

The representational + algorithmic baseline pair for v2 ByteDMD instrumentation. Pure numpy, laptop-runnable, paper-comparison metrics per stub:

• hinton-problems — 53 stubs from Hinton's papers (1981–2022). Site · Build notes.
• schmidhuber-problems — 58 stubs from Schmidhuber's papers (1989–2025). Site · Build notes.

How the cybertronai org repos fit together¶

graph LR
    ByteDMD["ByteDMD"]
    SimpleDally["simplified-
dally-model"]
    SPChall["sparse-parity-
challenge"]
    Hinton["hinton-problems"]
    Schmid["schmidhuber-
problems"]
    SutroP["sutro-problems"]
    Sutro["sutro"]
    SY["SutroYaro
(this repo)"]

    ByteDMD --> SY
    SimpleDally --> SY
    SPChall --> SY
    Hinton --> SY
    Schmid --> SY
    SutroP --> SY
    Sutro --> SY
    Sutro -.-> SPChall

    classDef cost fill:#e8d5ff,stroke:#7e3ff2,stroke-width:2px,color:#000
    classDef problem fill:#cfe5ff,stroke:#1f6feb,stroke-width:2px,color:#000
    classDef lab fill:#d1f4d1,stroke:#1a7f37,stroke-width:3px,color:#000

    class ByteDMD,SimpleDally cost
    class SPChall,Hinton,Schmid,SutroP,Sutro problem
    class SY lab

Purple = cost-metric repos · Blue = problem repos · Green = the lab notebook (this repo). Full curated map with descriptions: docs/related-repos.md.