Dontopedia

RotationalAdamW

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

RotationalAdamW has 28 facts recorded in Dontopedia across 12 references, with 3 live disagreements.

28 facts·20 predicates·12 sources·3 in dispute

Mostly:rdf:type(3), stores(2), not same as(2)

Maturity scale raw canonical shape-checked rule-derived certified

Inbound mentions (5)

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.

derivesMetricsFromDerives Metrics From(1)

identifiesTopPerformerIdentifies Top Performer(1)

involvesComponentInvolves Component(1)

requiresRequires(1)

usesComponentUses Component(1)

Other facts (24)

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.

24 facts
PredicateValueRef
Rdf:typeSoftware Component[8]
Rdf:typeOptimizer[10]
Rdf:typeOptimizer[11]
StoresSecond Moment V[3]
StoresFirst Moment M[3]
Not Same AsAdamw[9]
Not Same AsAdam With Weight Decay[9]
Not Required forModel Components[1]
Is Complextrue[1]
Is Optimizer Portdeferred[1]
Distinguished FromAdamw[2]
Stores PerParameter[3]
Teleologically OptimizesAdamw Variant[3]
Has Phase DynamicsWorking As Theory Predicts[4]
Exhibits Selective Crystallizationnull[4]
OptimizesModel[5]
Has Rot Strength0.1[5]
Contrasts WithAdam[6]
Provides Sphere Projection DiagnosticsSphere Projection Diagnostics[6]
Proposed As Stability Fixnull[7]
Has Performance ContextGPT-2 scale[8]
Needs Ternary Projection Steptrue[10]
Described Ascheap[11]
Has DiagnosticSphere Projection Diagnostics[12]

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.

28 undated
—
notRequiredForblah/watt-activation/part-102
ex:model-components
—
isComplexblah/watt-activation/part-102
true
—
isOptimizerPortblah/watt-activation/part-102
deferred
—
distinguishedFromblah/watt-activation/part-122
ex:adamw
—
storesblah/watt-activation/part-143
ex:second-moment-v
—
storesPerblah/watt-activation/part-143
ex:parameter
—
storesblah/watt-activation/part-143
ex:first-moment-m
—
teleologicallyOptimizesblah/watt-activation/part-143
ex:adamw-variant
—
hasPhaseDynamicsblah/watt-activation/part-188
ex:working-as-theory-predicts
—
exhibitsSelectiveCrystallizationblah/watt-activation/part-188
null
—
optimizesblah/watt-activation/part-305
ex:model
—
hasRotStrengthblah/watt-activation/part-305
0.1
—
contrastsWithblah/watt-activation/part-394
ex:adam
—
providesSphereProjectionDiagnosticsblah/watt-activation/part-394
ex:sphere-projection-diagnostics
—
proposedAsStabilityFixblah/watt-activation/part-609
null
—
typeblah/watt-activation/119
ex:SoftwareComponent
—
labelblah/watt-activation/119
RotationalAdamW
—
hasPerformanceContextblah/watt-activation/119
GPT-2 scale
—
notSameAsblah/watt-activation/122
ex:adamw
—
notSameAsblah/watt-activation/122
ex:adam-with-weight-decay
—
typeblah/watt-activation/275
ex:Optimizer
—
labelblah/watt-activation/275
RotationalAdamW
—
needsTernaryProjectionStepblah/watt-activation/275
true
—
labelblah/watt-activation/291
RotationalAdamW
—
typeblah/watt-activation/291
ex:Optimizer
—
describedAsblah/watt-activation/291
cheap
—
labelblah/watt-activation/392
RotationalAdamW
—
hasDiagnosticblah/watt-activation/392
ex:sphere-projection-diagnostics

References (12)

12 references
  1. [1]Part 1023 facts
    ctx:discord/blah/watt-activation/part-102
  2. [2]Part 1221 fact
    ctx:discord/blah/watt-activation/part-122
  3. [3]Part 1434 facts
    ctx:discord/blah/watt-activation/part-143
  4. [4]Part 1882 facts
    ctx:discord/blah/watt-activation/part-188
  5. [5]Part 3052 facts
    ctx:discord/blah/watt-activation/part-305
  6. [6]Part 3942 facts
    ctx:discord/blah/watt-activation/part-394
  7. [7]Part 6091 fact
    ctx:discord/blah/watt-activation/part-609
  8. [8]1193 facts
    ctx:discord/blah/watt-activation/119
    • full textwatt-activation-119
      text/plain3 KBdoc:agent/watt-activation-119/dd015076-4b38-4017-9483-3f91bdce858d
      Show excerpt
      [2026-03-09 00:25] xenonfun: okay at least generating something probablt still some bugs. ⏺ Committed and pushed. Key things done this session: 1. docs/symbiogenesis.md saved as a core document, linked prominently from CLAUDE.md 2. Roo…
  9. [9]1222 facts
    ctx:discord/blah/watt-activation/122
    • full textwatt-activation-122
      text/plain3 KBdoc:agent/watt-activation-122/57649dd0-cec5-4d9a-bc09-bec5f2db2137
      Show excerpt
      [2026-03-09 01:19] xenonfun: ⏺ BP = Backpropagation — whether the optimizer computes gradients via automatic differentiation or not. Adam / RotAdamW use standard backprop: 1. Forward pass → compute loss 2. nn.value_and_grad() → autod…
  10. [10]2753 facts
    ctx:discord/blah/watt-activation/275
    • full textwatt-activation-275
      text/plain3 KBdoc:agent/watt-activation-275/149bf24b-54fb-4412-b289-e2e03bccffe2
      Show excerpt
      [2026-03-13 20:46] xenonfun: ⏺ This tells a clear story: Current trained weights are NOT naturally ternary. Only ~46% of values are near {-1, 0, 1} after scaling (random Gaussian would give ~47%, so no better than chance). Relative err…
  11. [11]2913 facts
    ctx:discord/blah/watt-activation/291
    • full textwatt-activation-291
      text/plain3 KBdoc:agent/watt-activation-291/883c968e-0b03-4e27-aca9-027dfc155696
      Show excerpt
      [2026-03-14 03:41] xenonfun: Why Keeping the Spherical Head Still Makes Sense (performace at 600K parm scale is effectively same as euclidian head) ``` Even if performance is the same, the spherical head is still the better design. Reasons…
  12. [12]3922 facts
    ctx:discord/blah/watt-activation/392
    • full textwatt-activation-392
      text/plain1 KBdoc:agent/watt-activation-392/ddbc1ecc-c3dc-4acc-b74a-05cca6e39186
      Show excerpt
      [2026-03-19 03:47] xenonfun: ``` ❯ do we not have concept of SNR and bandwidth in these or they pure rotation and geometry? ⏺ They're pure rotation and geometry right now — no SNR or bandwidth concepts. The ResonantWireLM has no attention,…

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.

↑