Dontopedia

Multiheadkanattention

From Dontopedia, the open, paraconsistent wiki. (Last updated 2026-05-02.)

Multiheadkanattention has 14 facts recorded in Dontopedia across 1 reference.

14 facts·14 predicates·1 sources

Mostly:applies causal mask(1), concats heads(1), has d head(1)

Maturity scale raw canonical shape-checked rule-derived certified

Inbound mentions (1)

Other subjects in dontopedia point AT this entity as a value. These are inverse relationships — e.g. "X motherOf this subject" — and answer questions the forward facts can't. Grouped by predicate.

plansImplementationPlans Implementation(1)

Other facts (14)

The long tail: predicates that appear too rarely to warrant their own section. Filter or scroll to find a specific one. Each row links to its source.

14 facts
PredicateValueRef
Applies Causal MaskAttention[1]
Concats HeadsOutput[1]
Has D Headd_model // n_heads[1]
Has Learnable Temperature Per Headtrue[1]
Has N Headsn_heads[1]
Has Same Param Count AsKuramotoattention[1]
Is Drop in ReplacementBenchmark[1]
Is New Classtrue[1]
Matches Kuramoto Param CountW_q + W_k + W_v + W_o[1]
Mixes ValuesAttention Output[1]
Normalizes AttentionWeights[1]
Performs Q Kv Projections Reshape[B, L, n_heads, d_head][1]
Performs Transpose[B, n_heads, L, d_head][1]
Uses Per Head Sigmoidsigmoid(QK^T / √d_head)[1]

Timeline

Timeline axis is valid_time — when each source says the fact was true in the world, not when Dontopedia learned about it. Retracted rows are kept for provenance; coloured stripes indicate the context kind.

appliesCausalMaskblah/watt-activation/part-68
ex:attention
concatsHeadsblah/watt-activation/part-68
ex:output
hasDHeadblah/watt-activation/part-68
d_model // n_heads
hasLearnableTemperaturePerHeadblah/watt-activation/part-68
true
hasNHeadsblah/watt-activation/part-68
n_heads
hasSameParamCountAsblah/watt-activation/part-68
ex:kuramotoattention
isDropInReplacementblah/watt-activation/part-68
ex:benchmark
isNewClassblah/watt-activation/part-68
true
matchesKuramotoParamCountblah/watt-activation/part-68
W_q + W_k + W_v + W_o
mixesValuesblah/watt-activation/part-68
ex:attention-output
normalizesAttentionblah/watt-activation/part-68
ex:weights
performsQKvProjectionsReshapeblah/watt-activation/part-68
[B, L, n_heads, d_head]
performsTransposeblah/watt-activation/part-68
[B, n_heads, L, d_head]
usesPerHeadSigmoidblah/watt-activation/part-68
sigmoid(QK^T / √d_head)

References (1)

1 references
  1. [1]Part 6814 facts
    ctx:discord/blah/watt-activation/part-68

See also

Keep researching

Missing something or suspicious of what's here? Kick off a research session — a Claude agent will investigate, cite its sources, and file new facts into a dedicated context you can review before accepting into the shared view.