YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
NanoMind — FSI Efficiency Proof
A 6,824-parameter micro-convolutional network that hits 95.7% on MNIST at 9.7 ms/sample on CPU. The efficiency demonstration of the FSI (Ferrell Synthetic Intelligence) sovereign, off-grid, private-first model stack.
Design (our own, from scratch)
Depthwise-separable convolutions + Squeeze-Excite + a gated linear head. Tiny param budget, fewest multiply-adds, strong accuracy. No pretrained backbone.
Efficiency metrics
| metric | value |
|---|---|
| params | 6,824 |
| MACs (multiply-adds) | 308,896 |
| test accuracy (MNIST) | 95.69% |
| latency (batch=1, CPU) | 9.7 ms/sample |
Use
from nanomind import NanoMind, count_params, count_macs
m = NanoMind(); m.load_state_dict(torch.load("nanomind_best.pt")); m.eval()
print("params", count_params(m), "macs", count_macs(m))
nanomind_ts.pt is a TorchScript export for a portable CPU runtime.
The FSI stack
NanoMind (efficiency) · EchoCoder (coding specialist) · Sovereign Research Analyst
(legal public-OSINT collaborator). See STRATEGY.md.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support