Dontopedia

WireLM

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

WireLM has 31 facts recorded in Dontopedia across 9 references, with 2 live disagreements.

31 facts·21 predicates·9 sources·2 in dispute

Mostly:rdf:type(5), uses encoding(2), uses mechanism(2)

Maturity scale raw canonical shape-checked rule-derived certified

Inbound mentions (12)

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.

appliesToApplies to(1)

asksAboutAsks About(1)

doesNotTransferToDoes Not Transfer to(1)

handlesHandles(1)

hasFundamentallyDifferentBlockStructureThanHas Fundamentally Different Block Structure Than(1)

isFfnInIs Ffn in(1)

isHandledLikeIs Handled Like(1)

isPartOfIs Part of(1)

potentiallyImprovesBpbPotentially Improves Bpb(1)

targetsWireLMTargets Wire Lm(1)

transferableToTransferable to(1)

transfersToTransfers to(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:typeSoftware System[5]
Rdf:typeModel[6]
Rdf:typeModel Architecture[7]
Rdf:typeModel[8]
Rdf:typeModel Architecture[9]
Uses EncodingOfdm Encoding[3]
Uses EncodingOfdm Encoding[7]
Uses MechanismGated Cumsum Monoid[7]
Uses MechanismCumsum Based Attention[7]
Receives Compiled GenerationCompiled Generation[1]
Wired IntoHarmonicmlx Trainer Py[2]
Integrates Into TrainerHarmonicmlx Trainer Py[2]
Is Compatible WithLohe V3 Style[3]
Uses AttentionCumsum Based Attention[3]
Achieves SpeedO(1)/token[4]
Uses Cache TypeAttention step() states[4]
Uses Generation MethodToken-by-token[4]
Param Count8500000[6]
Uses ParameterGamma Parameter[7]
Gate CharacteristicFixed Gate[7]
Could Have FeatureInput Dependent Gates[7]
Decoder TypeSymmetric Decoder[7]
Currently Uses LayerLinear D Model 256[7]
Total Parameter Count8500000[7]
Proposed ChangeReplace Lohe Complex Split Attention With Rotor Blocks[7]
Uses Attention TypeLohe Complex Split[9]
Uses Ffn TypeLohe V3[9]

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.

31 undated
—
receivesCompiledGenerationblah/watt-activation/part-331
ex:compiled-generation
—
wiredIntoblah/watt-activation/part-329
ex:harmonicmlx-trainer-py
—
integratesIntoTrainerblah/watt-activation/part-329
ex:harmonicmlx-trainer-py
—
isCompatibleWithblah/watt-activation/part-405
ex:lohe-v3-style
—
usesEncodingblah/watt-activation/part-405
ex:ofdm-encoding
—
usesAttentionblah/watt-activation/part-405
ex:cumsum-based-attention
—
achievesSpeedblah/watt-activation/part-486
O(1)/token
—
usesCacheTypeblah/watt-activation/part-486
Attention step() states
—
usesGenerationMethodblah/watt-activation/part-486
Token-by-token
—
typeblah/watt-activation/329
ex:SoftwareSystem
—
labelblah/watt-activation/329
WireLM
—
typeblah/watt-activation/386
ex:Model
—
labelblah/watt-activation/386
WireLM
—
paramCountblah/watt-activation/386
8500000
—
typeblah/watt-activation/403
ex:ModelArchitecture
—
usesMechanismblah/watt-activation/403
ex:gated-cumsum-monoid
—
usesParameterblah/watt-activation/403
ex:gamma-parameter
—
gateCharacteristicblah/watt-activation/403
ex:fixed-gate
—
couldHaveFeatureblah/watt-activation/403
ex:input-dependent-gates
—
decoderTypeblah/watt-activation/403
ex:symmetric-decoder
—
usesEncodingblah/watt-activation/403
ex:ofdm-encoding
—
currentlyUsesLayerblah/watt-activation/403
ex:linear-d-model-256
—
totalParameterCountblah/watt-activation/403
8500000
—
usesMechanismblah/watt-activation/403
ex:cumsum-based-attention
—
proposedChangeblah/watt-activation/403
ex:replace-lohe-complex-split-attention-with-rotor-blocks
—
typeblah/watt-activation/422
ex:Model
—
labelblah/watt-activation/422
WireLM
—
typeblah/watt-activation/423
ex:ModelArchitecture
—
labelblah/watt-activation/423
WireLM
—
usesAttentionTypeblah/watt-activation/423
ex:lohe-complex-split
—
usesFFNTypeblah/watt-activation/423
ex:lohe-v3

References (9)

9 references
  1. [1]Part 3311 fact
    ctx:discord/blah/watt-activation/part-331
  2. [2]Part 3292 facts
    ctx:discord/blah/watt-activation/part-329
  3. [3]Part 4053 facts
    ctx:discord/blah/watt-activation/part-405
  4. [4]Part 4863 facts
    ctx:discord/blah/watt-activation/part-486
  5. [5]3292 facts
    ctx:discord/blah/watt-activation/329
    • full textwatt-activation-329
      text/plain2 KBdoc:agent/watt-activation-329/9423e8b2-3f97-4dd5-9105-db324a8b0017
      Show excerpt
      [2026-03-15 05:09] xenonfun: training now compiled: ⏺ Solo GPU numbers: ``` ┌────────────┬─────────┬────────┐ │ │ ms/step │ tok/s │ ├────────────┼─────────┼────────┤ │ Compiled │ 45.3 │ 90,380 │ ├────────────┼────…
  6. [6]3863 facts
    ctx:discord/blah/watt-activation/386
    • full textwatt-activation-386
      text/plain2 KBdoc:agent/watt-activation-386/56c807c0-0221-454e-97c5-567cf059cb84
      Show excerpt
      [2026-03-19 02:19] xenonfun: ``` ⏺ Sweep running — 7 configs sequential, no contention. ~15-20 min total. Configs: ┌─────┬───────────────┬────────┬──────────┬─────────┬───────────┬─────────────┐ │ # │ Config │ Params │ Dynamic…
  7. [7]40311 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…
  8. [8]4222 facts
    ctx:discord/blah/watt-activation/422
    • full textwatt-activation-422
      text/plain3 KBdoc:agent/watt-activation-422/a495a8d5-2cf4-41be-a842-8a6160f7c013
      Show excerpt
      [2026-03-20 01:24] xenonfun: Flaw 4: Optimizer micromanaging — INTENTIONALLY ADDRESSED (different fix) This one is nuanced. CLAUDE.md says: "Coupling gradient updates REMOVED: K and log_adjacency are structural constants, not learned." T…
  9. [9]4234 facts
    ctx:discord/blah/watt-activation/423
    • full textwatt-activation-423
      text/plain2 KBdoc:agent/watt-activation-423/8045fe79-be14-40ef-8627-1f24b38b2ba2
      Show excerpt
      [2026-03-20 01:28] xenonfun: ``` ⏺ OK, so for H≠4 it dispatches to _spectral_lohe_sync in attention.py (line 162), and for H=4 it uses quaternion_lohe_sync. Our entity run uses H=2, so it wouldn't even hit the Lohe FFN path — ResonantWir…

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.

↑