attention_weights — LLMCORE `score` op

• Data kinds: attnmap → attnmap

• Call: import fullseye as fs; fs.ledger.attention_weights(scores, mask=None, window=None) -> 'np.ndarray' (to call the implementation directly, import llmcore; llmcore.attention_weights(scores, mask=None, window=None) -> 'np.ndarray'; from the registry, opsllmcore.get("attention_weights"))

Usage

> This operator's description has not been translated yet. The original text follows as it is.

行ごとの softmax(マスクつき)→ attnmap。行和は 1、mask 位置は厳密に 0。

行ごとに最大値を引いてから指数を取る。これは飾りではない —— 引かずに `exp` すると

スコアが大きいとき全行が inf/NaN になる(実測: 96 行中 96 行)。引けば 0 行。

Args:

scores: (T, T) float の生スコア(`attention_scores` の出力)。

mask: None(全対)/ `"causal"(下三角)/ "window"(幅 window` の因果窓)/

(T, T) の bool 配列。True が「見てよい」。

window: `mask="window"` のときの窓幅(自分を含む個数)。

Returns:

attnmap (T, T) float: 各行が確率分布(和 1)。mask が False の位置は厳密に 0。

Detailed usage guide

• llmcore family guide

References (sample data, literature)

• Sample-data catalog (download URLs / licences) — 2-D uses skimage.data (BSD/public domain) plus synthetic images; 3-D lists download URLs for real data sources (Stanford, PDS, …).

• Operator provenance and references — the sources of the research/methods this op family came from.

• The canonical algorithm (author, year) and its uses are named in the family usage guide above.

Runnable examples (verified samples that actually call this op)

• poc_attention_identities — py -3.11 examples/poc_attention_identities.py

Ops the type connects to (they accept `attnmap` as input)

attention_apply

Same category (`score`)

attention_scores · attention_apply

*Provenance: llmcore.py — LLMCORE operator registry. This per-op note is generated by tools/opdocs.py md (do not hand-edit).*

attention_weights — LLMCORE score op