Dontopedia

Muon

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

Muon has 23 facts recorded in Dontopedia across 4 references, with 2 live disagreements.

23 facts·20 predicates·4 sources·2 in dispute

Mostly:cost amortization reason(2), per step cost context(2), produces nan(1)

Maturity scale raw canonical shape-checked rule-derived certified

Inbound mentions (2)

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.

componentOfComponent of(1)

isFixedTimeIs Fixed Time(1)

Other facts (22)

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.

22 facts
PredicateValueRef
Cost Amortization Reasonoptimizer step is fixed-time[4]
Cost Amortization Reasonforward/backward grow with T[4]
Per Step Cost Contextlong-context training regime[4]
Per Step Cost Contextarchitecture shines[4]
Produces NanBy Step 100[1]
Compared toAdam W[2]
Superior in Long ContextAdam W[2]
Known for Fixed Overhead{}[2]
Costs Throughput15-30%[2]
Has Params119.358M[3]
Has Per Layer Groups57[3]
Rdf:typeOptimizer[4]
Cost Comparison VsAdamw[4]
Throughput Cost Lower Bound15[4]
Throughput Cost Upper Bound30[4]
Throughput Cost Unitpercent[4]
Cost ReasonNS5 iteration on each 2D weight[4]
Cost Amortized at Long Ttrue[4]
Overhead Typefixed cost[4]
Overhead Impact Conditioneach step is short[4]
Per Step Cost Relativenoise relative to AdamW[4]
Per Step Cost ConditionT ≥ 2K[4]

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.

producesNanblah/watt-activation/part-694
ex:by-step-100
comparedToblah/watt-activation/part-693
ex:adam-w
superiorInLongContextblah/watt-activation/part-693
ex:adam-w
knownForFixedOverheadblah/watt-activation/part-693
{}
costsThroughputblah/watt-activation/part-693
15-30%
hasParamsblah/watt-activation/part-699
119.358M
hasPerLayerGroupsblah/watt-activation/part-699
57
labelblah/watt-activation/690
Muon
typeblah/watt-activation/690
ex:Optimizer
costComparisonVsblah/watt-activation/690
ex:adamw
throughputCostLowerBoundblah/watt-activation/690
15
throughputCostUpperBoundblah/watt-activation/690
30
throughputCostUnitblah/watt-activation/690
percent
costReasonblah/watt-activation/690
NS5 iteration on each 2D weight
costAmortizedAtLongTblah/watt-activation/690
true
costAmortizationReasonblah/watt-activation/690
optimizer step is fixed-time
costAmortizationReasonblah/watt-activation/690
forward/backward grow with T
overheadTypeblah/watt-activation/690
fixed cost
overheadImpactConditionblah/watt-activation/690
each step is short
perStepCostRelativeblah/watt-activation/690
noise relative to AdamW
perStepCostConditionblah/watt-activation/690
T ≥ 2K
perStepCostContextblah/watt-activation/690
long-context training regime
perStepCostContextblah/watt-activation/690
architecture shines

References (4)

4 references
  1. [1]Part 6941 fact
    ctx:discord/blah/watt-activation/part-694
  2. [2]Part 6934 facts
    ctx:discord/blah/watt-activation/part-693
  3. [3]Part 6992 facts
    ctx:discord/blah/watt-activation/part-699
  4. [4]69016 facts
    ctx:discord/blah/watt-activation/690
    • full textwatt-activation-690
      text/plain1 KBdoc:agent/watt-activation-690/506c50ab-67a0-4d6a-95fd-dbf8de71ca9e
      Show excerpt
      [2026-04-28 09:25] xenonfun: Noted — designer says Muon-manifold is the highest-impact lever. That's consistent with the harmonicrust ecosystem: wave_unified_muon_train already uses Muon (NS5) on proj_out + ManifoldMuon on omega + Rotationa

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.