Skip to content
SUNSETThe commons
NEAR TESTNET
THE OPEN KNOWLEDGE COMMONS

Useful ideas.
Yours to keep.

Original technical guides. Worked examples. Real calculations. No wallet to read.

TRANSFORMERSsoftmax(QKᵀ / √dₖ)V01 / OPEN RESOURCE
CC0 · FREE TO READ18 min

Transformer Attention Lab

Work through a complete attention head: tensor shapes, scaled logits, stable softmax, causal masking and the cost of a longer context. Original examples for readers comfortable with vectors.

Read & explore
NEURAL NETWORKS∂L/∂w = (p − y)x02 / OPEN RESOURCE
CC0 · FREE TO READ16 min

Backpropagation Field Notes

Follow a binary classifier from a forward pass to its exact gradients, one parameter update and a finite-difference check. A compact derivation you can reproduce on paper.

Read & explore
EFFICIENT ADAPTATIONΔW = (α/r)BA03 / OPEN RESOURCE
CC0 · FREE TO READ15 min

Low-Rank Adaptation Notebook

Derive a LoRA update, count the trainable parameters and weight bytes, and test the rank constraint with an explicit matrix. Separate useful savings from total-memory claims.

Read & explore
THE RELEASE LAYER

Backed by actual chain state.

Wallet actions use the existing NEAR testnet contract. New campaigns are curated by its creator account.

Explore onchain campaigns ↗
Reading NEAR testnet…