Dontopedia

LoheFFNv3

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

LoheFFNv3 has 33 facts recorded in Dontopedia across 10 references, with 4 live disagreements.

33 facts·18 predicates·10 sources·4 in dispute

Mostly:rdf:type(5), computes feature(4), implementation change(3)

Maturity scale raw canonical shape-checked rule-derived certified

Inbound mentions (7)

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.

connectsConnects(1)

hasComponentHas Component(1)

modifiesModifies(1)

mostlyRedundantWithMostly Redundant With(1)

portsPorts(1)

sameDecompositionAsSame Decomposition As(1)

usesUses(1)

Other facts (27)

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.

27 facts
PredicateValueRef
Rdf:typeSynchronization Method[4]
Rdf:typeOperation[6]
Rdf:typeFeed Forward Network[7]
Rdf:typeArchitecture[8]
Rdf:typeModel Version[9]
Computes FeatureRaw Synced Spectra[9]
Computes FeatureDft Ring Eigenmodes[9]
Computes FeatureR Global[9]
Computes FeatureJacobi Polynomials[9]
Implementation ChangeAdd K Probes Param[9]
Implementation ChangeAdd Probe Weights[9]
Implementation ChangeAppend Features[9]
Refers to ArchitectureIntra Block Ring Sync[1]
SyncsRing[2]
Has Ring Graph DesignTrue[3]
Classificationsparse graph[4]
Uses Decomposition Within BlockDft Modes[5]
Operates WithinEach Block[5]
Has Duration1.5[6]
Has Percentage of Block8[6]
Is More Parameter Efficient ThanStandard Transformer Ffn[7]
Has DynamicsV3 Dynamics[9]
Estimated Code Change Size~10 lines of change[9]
Parameter Count Increase256[9]
Latency Increase~0.2ms extra[9]
Has FeatureClifford Readout[10]
FlagClifford Readout Flag[10]

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.

refersToArchitectureblah/watt-activation/part-226
ex:intra-block-ring-sync
syncsblah/watt-activation/part-305
ex:ring
hasRingGraphDesignblah/watt-activation/part-313
ex:true
typeblah/watt-activation/225
ex:SynchronizationMethod
labelblah/watt-activation/225
LoheFFNv3
classificationblah/watt-activation/225
sparse graph
labelblah/watt-activation/228
LoheFFNv3
usesDecompositionWithinBlockblah/watt-activation/228
ex:dft-modes
operatesWithinblah/watt-activation/228
ex:each-block
labelblah/watt-activation/291
LoheFFNv3
typeblah/watt-activation/291
ex:Operation
hasDurationblah/watt-activation/291
1.5
hasPercentageOfBlockblah/watt-activation/291
8
labelblah/watt-activation/301
LoheFFNv3
typeblah/watt-activation/301
ex:FeedForwardNetwork
isMoreParameterEfficientThanblah/watt-activation/301
ex:standard-transformer-ffn
typeblah/watt-activation/318
ex:Architecture
labelblah/watt-activation/318
LoheFFNv3
typeblah/watt-activation/380
ex:ModelVersion
labelblah/watt-activation/380
LoheSphericalFFNv3
hasDynamicsblah/watt-activation/380
ex:v3-dynamics
computesFeatureblah/watt-activation/380
ex:raw-synced-spectra
computesFeatureblah/watt-activation/380
ex:dft-ring-eigenmodes
computesFeatureblah/watt-activation/380
ex:r-global
computesFeatureblah/watt-activation/380
ex:jacobi-polynomials
implementationChangeblah/watt-activation/380
ex:add-k-probes-param
implementationChangeblah/watt-activation/380
ex:add-probe-weights
implementationChangeblah/watt-activation/380
ex:append-features
estimatedCodeChangeSizeblah/watt-activation/380
~10 lines of change
parameterCountIncreaseblah/watt-activation/380
256
latencyIncreaseblah/watt-activation/380
~0.2ms extra
hasFeatureblah/watt-activation/403
ex:clifford-readout
flagblah/watt-activation/403
ex:clifford-readout-flag

References (10)

10 references
  1. [1]Part 2261 fact
    ctx:discord/blah/watt-activation/part-226
  2. [2]Part 3051 fact
    ctx:discord/blah/watt-activation/part-305
  3. [3]Part 3131 fact
    ctx:discord/blah/watt-activation/part-313
  4. [4]2253 facts
    ctx:discord/blah/watt-activation/225
    • full textwatt-activation-225
      text/plain2 KBdoc:agent/watt-activation-225/9b15ed6a-4fd9-4458-9e17-1e0372b5a3e0
      Show excerpt
      [2026-03-11 05:15] xenonfun: ⏺ Ablation complete. The result is more interesting than "hub wins": blk acc lift note 0 0.575 +0.217 ← input (BEST) 4 0.567 +0.208 3 0.558 +0.200 7 0.550 +0.192 ... 9
  5. [5]2283 facts
    ctx:discord/blah/watt-activation/228
    • full textwatt-activation-228
      text/plain3 KBdoc:agent/watt-activation-228/218c4ece-61aa-47aa-a16e-b77f7f994d5f
      Show excerpt
      [2026-03-11 05:23] xenonfun: ⏺ It is, and the math backs it up cleanly. What the network discovered spontaneously is essentially a depth-wise DFT — the same decomposition that LoheFFNv3 uses within each block (ring → DFT modes), but now o
  6. [6]2914 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
  7. [7]3013 facts
    ctx:discord/blah/watt-activation/301
    • full textwatt-activation-301
      text/plain3 KBdoc:agent/watt-activation-301/3e98420c-16db-4c57-9cad-a987cec9976b
      Show excerpt
      [2026-03-14 06:51] xenonfun: ``` ⏺ Lohe vs Softmax AR Decoder Comparison ┌───────────────┬───────────────┬────────────┐ │ Metric │ Softmax d=384 │ Lohe d=384 │ ├───────────────┼───────────────┼────────────┤ │ BPB │
  8. [8]3182 facts
    ctx:discord/blah/watt-activation/318
    • full textwatt-activation-318
      text/plain3 KBdoc:agent/watt-activation-318/f52d95a8-f461-40d1-9360-f08558b18eb1
      Show excerpt
      [2026-03-15 02:47] xenonfun: ⏺ I see you're working on wire encoding / phase modulation — that's a fascinating direction. Let me check what you've got: [2026-03-15 02:47] lisamegawatts: Wire QPSK + Standard: PPL 4.94, Byte Accuracy 51.5% T
  9. [9]38013 facts
    ctx:discord/blah/watt-activation/380
    • full textwatt-activation-380
      text/plain3 KBdoc:agent/watt-activation-380/84a046b5-55b5-4f12-a277-387a85fc8228
      Show excerpt
      [2026-03-19 00:07] xenonfun: ``` ❯ would we want to make a LoheFFNv4 that just replaces that readout or does that add more to it? ⏺ Just the readout. Here's why: v3 dynamics are sound — ring sync, DFT modes, Lohe tangent step all work.
  10. [10]4032 facts
    ctx:discord/blah/watt-activation/403
    • full textwatt-activation-403
      text/plain2 KBdoc:agent/watt-activation-403/d55f91e3-d6ae-47d0-a701-fcf659140c12
      Show excerpt
      [2026-03-19 05:31] xenonfun: ❯ is any of this applicable to WireLM lohe_v3 style stuff? ``` ⏺ Yes — the decay gates specifically. The gated cumsum monoid (g, v) ⊗ (g', v') = (g·g', g'·v + v') is exactly what _gated_cumsum already does in at

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.