MiniEmbed-Nano / README.md

betterwithage

MiniEmbed-Nano card

940d794 verified about 23 hours ago

preview code

Raw

History Blame Contribute Delete

2.35 kB

metadata

license: apache-2.0
library_name: numpy
tags:
  - governed-ai
  - khipu
  - szl-holdings
  - embedding
  - silhouette

MiniEmbed-Nano

Tiny hash+table embed: V=64, d=12, L2-normalized rows. Not a foundation embed. Not neural. Not MiniEmbed 3290×128.

Canonical source: szl-holdings/szl-khipu
Sibling card: SZLHOLDINGS/szl-khipu
The larger statistical MiniEmbed (3290 × 128) lives on SZLHOLDINGS/szl-kernels — a different artifact. Do not mix them.

from szl_khipu.train import mini_embed

emb = mini_embed.build(seed=20260721)
vec = emb.embed("knot the run")
print(emb.V, emb.D, vec.shape)
# 64 12 (12,)
emb.save_npz("mini_embed.npz")

What it does

SHA-256 token id modulo 64. Mean-pool then L2. Deterministic given seed.
Built here on CPU NumPy. Honesty REPORTED. Energy UNAVAILABLE.
No analogy score. No retrieval score. No SVD variance claim (that belongs to the 3290×128 table).

Bench (this tree)

TRAINING_RECEIPT.json seed 20260721 · honesty REPORTED

Metric	Value
V×d	64 × 12
method	hash+table L2
weights	`mini_embed.npz` sha256 `ae31a3a7214d1f142d8ea3f4f86c35bdedd7c108bc5d04ea00c87e7b674e6e3b`

Infers on POST /api/infer {"kind":"mini_embed","token":"F18"}. Not neural. Not 3290×128.

What it is NOT

Not the SZLHOLDINGS/szl-kernels MiniEmbed (3290 × 128, SVD var 0.3146).
Not neural. Not word2vec. Not a foundation embed.
Not 1.5B. Not Qwen.
Not proven trust. Λ uniqueness remains Conjecture 1 OPEN.
Energy UNAVAILABLE. CUDA UNAVAILABLE. Never a fabricated joule.

Honesty

Claim	Label	What-NOT
Table built in this package	REPORTED	V=64 d=12, not 3290×128
Neural / trained embed	FALSE	hash+table, not SGD
Analogy / retrieval score	UNAVAILABLE	not measured
Energy	UNAVAILABLE	never a fabricated joule
CUDA	UNAVAILABLE	CPU numpy LIVE