Dontopedia
Explore

Kan Attention Swap

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

Kan Attention Swap has 24 facts recorded in Dontopedia across 1 reference, with 1 live disagreement.

24 facts·22 predicates·1 sources·1 in dispute

Mostly:is faster by(3), achieves o1 step inference(1), breaks lohe structure(1)

Maturity scale raw canonical shape-checked rule-derived certified

Is Faster byin disputeisFasterBy

  • 7.5x[1]all time · Part 647
  • 33x[1]all time · Part 647
  • 265x[1]all time · Part 647

Achieves O1 Step InferenceachievesO1StepInference

Breaks Lohe StructurebreaksLoheStructure

Comparable Effort tocomparableEffortTo

Compared tocomparedTo

Evaluated Negatively on ParamsevaluatedNegativelyOnParams

Evaluated Positively on SpeedevaluatedPositivelyOnSpeed

Has Forward ComplexityhasForwardComplexity

  • O(T)[1]all time · Part 647

Has Unknown QualityhasUnknownQuality

Has VerdicthasVerdict

  • Future experiment, not for the 50M run[1]all time · Part 647

Increases Parameters byincreasesParametersBy

  • 99%[1]all time · Part 647

Lacks Backward PasslacksBackwardPass

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.

hasKeyFindingsHas Key Findings(1)

Other facts (10)

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.

10 facts
PredicateValueRef
Lacks Python EvidenceTrue[1]
Mismatches Group ConventionTrue[1]
Needs Python ValidationTrue[1]
Requires Halving LayersTrue[1]
Requires Implementationforward_train(), KANAttentionCache, analytical backward[1]
Requires Loc800-1200[1]
Scales LinearlyTrue[1]
Uses Projections4[1]
Is Inference Only If FailsTrue[1]
Suitable for Long ContextTrue[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.

achievesO1StepInferenceblah/watt-activation/part-647
ex:true
breaksLoheStructureblah/watt-activation/part-647
ex:true
comparableEffortToblah/watt-activation/part-647
ex:option-b
comparedToblah/watt-activation/part-647
ex:lohe-attention
evaluatedNegativelyOnParamsblah/watt-activation/part-647
ex:true
evaluatedPositivelyOnSpeedblah/watt-activation/part-647
ex:true
hasForwardComplexityblah/watt-activation/part-647
O(T)
hasUnknownQualityblah/watt-activation/part-647
ex:true
hasVerdictblah/watt-activation/part-647
Future experiment, not for the 50M run
increasesParametersByblah/watt-activation/part-647
99%
isFasterByblah/watt-activation/part-647
7.5x
isFasterByblah/watt-activation/part-647
33x
isFasterByblah/watt-activation/part-647
265x
isInferenceOnlyIfFailsblah/watt-activation/part-647
ex:true
lacksBackwardPassblah/watt-activation/part-647
ex:true
lacksPythonEvidenceblah/watt-activation/part-647
ex:true
mismatchesGroupConventionblah/watt-activation/part-647
ex:true
needsPythonValidationblah/watt-activation/part-647
ex:true
requiresHalvingLayersblah/watt-activation/part-647
ex:true
requiresImplementationblah/watt-activation/part-647
forward_train(), KANAttentionCache, analytical backward
requiresLOCblah/watt-activation/part-647
800-1200
scalesLinearlyblah/watt-activation/part-647
ex:true
suitableForLongContextblah/watt-activation/part-647
ex:true
usesProjectionsblah/watt-activation/part-647
4

References (1)

1 references
  1. [1]Part 64724 facts
    customctx:discord/blah/watt-activation/part-647

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.