Python · AGPL-3.0
Scottcjn/ram-coffers
LLM infrastructure cost reduction via NUMA-aware weight banking: 147 t/s (8.8x stock llama.cpp) on refurbished enterprise POWER8. Self-hosted inference, no cloud APIs. Part of the Proof of Physical AI stack.
★ 173
Stars
59 forks · 31 open issues
+2.0/wk
Growth · measured
Stars counted between two of our own snapshots roughly a week apart. The repository page gives the exact span.
20.4
Momentum / 100
see the breakdown below
9mo ago
Created
last push 41d ago
Star history
How this score was built
| Component | Weight | Earned | Why |
|---|---|---|---|
| Velocity | 55 | 2.1 | +2.0/wk, log-scaled against a 750/day ceiling |
| Acceleration | 15 | 0.0 | needs 7-day and 30-day rates side by side |
| Freshness | 15 | 11.8 | last push 41d ago; decays to zero at 180 days |
| Scale | 15 | 6.5 | 173 stars, log-scaled against 150k — stops a 3-star repo topping the board |
| Momentum | 100 | 20.4 | sum of the above |
Detail
| Topics | ai-inferenceai-infrastructurecost-optimizationdepinedge-aienergy-efficiencyhebbianinference-optimizationllama-cppllmllm-inferencellmopsnumapower8powerpcppc64leproof-of-physical-aiself-hostedtransformervec-perm |
| Matched on | LLM & Inference (llm-inference) |
| Short window | +2 stars over 7 days = 0.3/day |
| Long window | no snapshot 14–45 days old yet |
| Lifetime average | 0.60 stars/day over 290 days |
| First seen here | 2026-10-29 |
| Last read | 2d ago |