Dontopedia

WireLM

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

WireLM has 27 facts recorded in Dontopedia across 5 references, with 2 live disagreements.

27 facts·24 predicates·5 sources·2 in dispute

Mostly:rdf:type(2), achieves percent better per byte information content(1), has bpb(1)

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.

isRunOfIs Run of(2)

appliesToModelApplies to Model(1)

extendsExtends(1)

hasModelHas Model(1)

hasTinyParamsRelativeToHas Tiny Params Relative to(1)

implementsModelImplements Model(1)

Other facts (25)

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.

25 facts
PredicateValueRef
Rdf:typeLanguage Model[4]
Rdf:typeModel[5]
Achieves Percent Better Per Byte Information Content8[1]
Has Bpb1.572[1]
Has Three Times Fewer Params ThanSpectral Plus Lohe V3 Bpe 8k[1]
Is Byte Level Modelnull[1]
Uses Quaternion Architecturenull[1]
Has Params Count8500000[2]
Serves As Size BaselineSweep[2]
Relies on Lohe Componentsattention + FFN[3]
Uses AttentionLohe Complex Split Attention[3]
Uses FfnLohe V3 Ffn[3]
Has Parameter Count8250000[4]
Measured Performance in Nats Per Byte1.09[4]
Trained for Steps20000[4]
Outperforms on MetricSpectral Lohe V3 Bpe 8k[4]
Has Fewer Parameters ThanSpectral Lohe V3 Bpe 8k[4]
Has Three Times Fewer Parameters ThanSpectral Lohe V3 Bpe 8k[4]
Has Improved Information Content by Percentage8[4]
Predicts Tokens Per Byte1[4]
Has Performance Slope-0.052[4]
Measured Performance in Bpb1.572[4]
Has Cache TypeAttention step() states[5]
Has Generation MethodToken-by-token[5]
Has SpeedO(1)/token[5]

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.

achievesPercentBetterPerByteInformationContentblah/watt-activation/part-327
8
hasBpbblah/watt-activation/part-327
1.572
hasThreeTimesFewerParamsThanblah/watt-activation/part-327
ex:spectral-plus-lohe-v3-bpe-8k
isByteLevelModelblah/watt-activation/part-327
null
usesQuaternionArchitectureblah/watt-activation/part-327
null
hasParamsCountblah/watt-activation/part-388
8500000
servesAsSizeBaselineblah/watt-activation/part-388
ex:sweep
reliesOnLoheComponentsblah/watt-activation/part-425
attention + FFN
usesAttentionblah/watt-activation/part-425
ex:lohe-complex-split-attention
usesFFNblah/watt-activation/part-425
ex:lohe-v3-ffn
labelblah/watt-activation/325
Wire LM quaternion
typeblah/watt-activation/325
ex:LanguageModel
hasParameterCountblah/watt-activation/325
8250000
measuredPerformanceInNatsPerByteblah/watt-activation/325
1.09
trainedForStepsblah/watt-activation/325
20000
outperformsOnMetricblah/watt-activation/325
ex:spectral-lohe-v3-bpe-8k
hasFewerParametersThanblah/watt-activation/325
ex:spectral-lohe-v3-bpe-8k
hasThreeTimesFewerParametersThanblah/watt-activation/325
ex:spectral-lohe-v3-bpe-8k
hasImprovedInformationContentByPercentageblah/watt-activation/325
8
predictsTokensPerByteblah/watt-activation/325
1
hasPerformanceSlopeblah/watt-activation/325
-0.052
measuredPerformanceInBpbblah/watt-activation/325
1.572
labelblah/watt-activation/484
WireLM
typeblah/watt-activation/484
ex:Model
hasCacheTypeblah/watt-activation/484
Attention step() states
hasGenerationMethodblah/watt-activation/484
Token-by-token
hasSpeedblah/watt-activation/484
O(1)/token

References (5)

5 references
  1. [1]Part 3275 facts
    ctx:discord/blah/watt-activation/part-327
  2. [2]Part 3882 facts
    ctx:discord/blah/watt-activation/part-388
  3. [3]Part 4253 facts
    ctx:discord/blah/watt-activation/part-425
  4. [4]32512 facts
    ctx:discord/blah/watt-activation/325
    • full textwatt-activation-325
      text/plain3 KBdoc:agent/watt-activation-325/2c3de55b-7643-47a7-9671-90e12c7ea3ba
      Show excerpt
      [2026-03-15 04:18] xenonfun: ``` ⏺ This is the real story: Wire LM (8.25M) is beating the best BPE model (24.47M) on per-byte CE: ┌─────────────────────────┬────────┬───────────┬───────┐ │ Model │ Params │ nats/byt
  5. [5]4845 facts
    ctx:discord/blah/watt-activation/484
    • full textwatt-activation-484
      text/plain2 KBdoc:agent/watt-activation-484/0f2365f0-484f-40bd-bbda-2a0aa92a1109
      Show excerpt
      [2026-03-22 01:01] xenonfun: ``` ⏺ Pushed. All models now have cached inference: ┌────────────────────┬───────────────────────────┬────────────────┬───────────────────┐ │ Model │ Cache Type │ Generation

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.