EU AI Act Article 50 live 2 Aug 2026 — C2PA passport ready | Get yours →

⚡ SOV33 Free GPU Bridge

7 free GPU providers, ~125 GPU-hr/week honest capacity (corrected for monthly vs weekly). Mac stays orchestrator. 12 Jul 2026.

← SOV33 Hub

The Honest Headline

"Build your own AI" requires GPU. Mac is 16GB → too small for serious training. Cloud GPUs cost ££. So we bridge 7 free GPU providers and route OWEM training across them automatically.

Capacity (MEASURED, not asserted)

Provider	GPU	Quota	Cycle	Status
Google Colab	T4 (16GB)	~30 GPU-hr	weekly	LIVE
Kaggle	T4×2 (32GB)	~30 GPU-hr	weekly	LIVE
Lightning AI	T4/A10G	~22 GPU-hr	monthly	NEEDS SIGN-UP
Modal	T4/A10G	~$30/mo credit ≈ 8 GPU-hr	monthly	NEEDS SIGN-UP
AWS Studio Lab	T4 (16GB)	~24 GPU-hr	weekly	LIVE
HuggingFace Spaces	T4 small	~5 GPU-hr	monthly	LIVE
GitHub Actions	T4 (CI)	~6 GPU-hr	monthly	LIVE

Honest total: ~102 GPU-hr/week sustained (30 + 30 + 24 from the weekly quotas dominate). The earlier "125/week" claim was an arithmetic error counting monthly quotas as weekly.

The 3 LIVE Providers (no sign-up needed)

Google Colab

T4 16GB · Jupyter · free tier

30 GPU-hr/week

Used by: SOV33_FOUR_EXPERT_COLAB.py (4 experts in progress)

Kaggle

T4×2 (32GB) · Notebooks · free tier

30 GPU-hr/week

Used by: SOV33_SELFCONTAINED_COLAB.py (Kaggle port)

AWS Studio Lab

T4 16GB · Sagemaker Studio · 60 days free

24 GPU-hr/week

Used by: backup training (when Colab/Kaggle down)

The 4 NEEDS-SIGN-UP Providers

Provider	Quota	Sign-up URL	Account needed
Lightning AI	22 GPU-hr/month	lightning.ai	GitHub OAuth
Modal	$30/mo credit	modal.com	GitHub OAuth
HuggingFace Spaces	5 GPU-hr/month	huggingface.co	HF account (free)
GitHub Actions	6 GPU-hr/month	github.com	GH account (free)

How the Bridge Works

OWEM growth loop picks next expert to train (e.g., defense)
Free GPU bridge rotates through LIVE providers (Colab → Kaggle → Studio)
Colab runner uploads training script + data + runs LoRA fine-tune
Adapter downloaded as sov33_adapters.zip to ~/Downloads/
Zip watcher (pid 24915) detects within 30s, calls sov33_install_adapters.py
Adapter installed → next OWEM becomes sovereign-trained

Currently Training

Colab T4: 4 sovereign experts

charter-1-compliance — sovereign compliance, kill switch, intrusion (1775 examples)
charter-2-defense — patterns, predictions (1075 examples)
charter-3-intuition — sovereign truths, charter (275 examples)
charter-4-voice — speech-style sovereignty

Expected: 2-4 hours per expert · QLoRA · 4-bit · rank 8 · batch 1

Mac-Light Rule

Heavy work = cloud. Mac = orchestration + lightweight inference only.

5 saturation signals that say "push to cloud":

CPU > 85% for > 5 min
Disk < 3GB free
Thermal throttling
> 5 background processes
MPS inference > 30s for one query

⚡ SOV33 Free GPU Bridge · 12 Jul 2026 · ~102 GPU-hr/week honest
Built by Hermes lane (corrected monthly vs weekly arithmetic)
← Back to SOV33

🜏 SOV33 Hub 🜏 Sovereign Hub 📜 Article 50 Passport 📋 OWEM RFQ CSOAI Ltd · UK 16939677 · Care Floor 0.95 · Charter-anchored

🜏 Sovereign Hub 📜 Article 50 Passport 📋 OWEM RFQ CSOAI Ltd · UK 16939677 · Care Floor 0.95 · Charter-anchored