universal

as donto believed on

From Dontopedia, the open, paraconsistent wiki. (Last updated 2026-06-06.)

universal has 45 facts recorded in Dontopedia across 4 references, with 3 live disagreements.

45 facts·36 predicates·4 sources·3 in dispute

Mostly:involves operations(5), has extra learnable vectors(2), rdf:type(2)

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.

`activationCompetitor`Activation Competitor(1)

Super Chat 20260225143238 Sa7hex:super-chat-20260225143238-sa7h

Other facts (43)

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.

43 facts

Predicate	Value	Ref
Involves Operations	Matmul	[4]
Involves Operations	Silu	[4]
Involves Operations	Mul Gate	[4]
Involves Operations	Mul Skip	[4]
Involves Operations	Add	[4]
Has Extra Learnable Vectors	2	[1]
Has Extra Learnable Vectors	2	[4]
Rdf:type	Activation Function	[2]
Rdf:type	Activation Function	[4]
Has Initial Config	gate=1	[4]
Has Initial Config	skip=0	[4]
Adapts Activation Shape Over Steps	20	[1]
Causes Lower Tok Per Second	Same Hardware	[1]
Does Not Get Faster	true	[1]
Evaluated Positively for Loss Drop	true	[1]
Ffn Forward Pass Sequence	matmul → silu → mul(gate) → mul(skip) → add → matmul	[1]
Has Fixed Tok Per Second	true	[1]
Has Lower Tok Per Second	Simple Activations	[1]
Has Op Count Per Ffn Forward Pass	6	[1]
Has Ops Multiplier Relative to Gelu	2	[1]
Improves Loss Per Compute	true	[1]
Improves Loss Per Step	true	[1]
Known to Have Gate and Skip Params	true	[1]
Learns Per Channel Blends	true	[1]
Optimizes Learnable Params Via Backprop	true	[1]
Selected Despite Slower Speed	Symbio Fitness	[1]
Starts As	Silu Activation	[1]
Wins If Drops Loss Faster	2× faster per step than gelu	[1]
Learnable Param Count	2	[2]
Definition	f(x) = silu(x)gate + xskip	[2]
Init Strategy	starts as SiLU	[2]
Learning Goal	blend nonlinear + linear paths per channel	[2]
Init Gate Value	1	[2]
Init Skip Value	0	[2]
Init Mode	pure SiLU	[2]
Param Change During Warmup	Minimal Movement	[3]
State During Warmup	Fixed Starting Point	[3]
Has Speed Over Time	Fixed Speed	[4]
Has Op Count	6	[4]
Has Learnable Vector Shape	[1,ffnDim]	[4]
Has Learnable Params	gate/skip	[4]
Initializes As	SiLU	[4]
Learns Feature	per-channel blends	[4]

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.

45 undated

—

adaptsActivationShapeOverStepsblah/training-and-evals/part-28

—

causesLowerTokPerSecondblah/training-and-evals/part-28

ex:same-hardware

—

doesNotGetFasterblah/training-and-evals/part-28

true

—

evaluatedPositivelyForLossDropblah/training-and-evals/part-28

true

—

ffnForwardPassSequenceblah/training-and-evals/part-28

matmul → silu → mul(gate) → mul(skip) → add → matmul

—

hasExtraLearnableVectorsblah/training-and-evals/part-28

—

hasFixedTokPerSecondblah/training-and-evals/part-28

true

—

hasLowerTokPerSecondblah/training-and-evals/part-28

ex:simple-activations

—

hasOpCountPerFfnForwardPassblah/training-and-evals/part-28

—

hasOpsMultiplierRelativeToGelublah/training-and-evals/part-28

—

improvesLossPerComputeblah/training-and-evals/part-28

true

—

improvesLossPerStepblah/training-and-evals/part-28

true

—

knownToHaveGateAndSkipParamsblah/training-and-evals/part-28

true

—

learnsPerChannelBlendsblah/training-and-evals/part-28

true

—

optimizesLearnableParamsViaBackpropblah/training-and-evals/part-28

true

—

selectedDespiteSlowerSpeedblah/training-and-evals/part-28

ex:symbio-fitness

—

startsAsblah/training-and-evals/part-28

ex:silu-activation

—

winsIfDropsLossFasterblah/training-and-evals/part-28

2× faster per step than gelu

—

typeblah/training-and-evals/23

ex:ActivationFunction

—

labelblah/training-and-evals/23

universal

—

learnableParamCountblah/training-and-evals/23

—

definitionblah/training-and-evals/23

f(x) = silu(x)*gate + x*skip

—

initStrategyblah/training-and-evals/23

starts as SiLU

—

learningGoalblah/training-and-evals/23

blend nonlinear + linear paths per channel

—

initGateValueblah/training-and-evals/23

—

initSkipValueblah/training-and-evals/23

—

initModeblah/training-and-evals/23

pure SiLU

—

paramChangeDuringWarmupblah/training-and-evals/24

ex:minimal-movement

—

stateDuringWarmupblah/training-and-evals/24

ex:fixed-starting-point

—

labelblah/training-and-evals/28

universal activation

—

typeblah/training-and-evals/28

ex:ActivationFunction

—

hasSpeedOverTimeblah/training-and-evals/28

ex:fixed-speed

—