Dontopedia

block-diagonal mode

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

block-diagonal mode has 9 facts recorded in Dontopedia across 2 references, with 1 live disagreement.

9 facts·7 predicates·2 sources·1 in dispute

Mostly:partitions residual stream into(2), causes loss of(1), loses feature(1)

Maturity scale raw canonical shape-checked rule-derived certified

Inbound mentions (1)

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.

providesWhatIsLostByProvides What Is Lost by(1)

Other facts (8)

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.

8 facts
PredicateValueRef
Partitions Residual Stream IntoG Equals 8 Independent 96 Dim Lanes[1]
Partitions Residual Stream Intoindependent lanes[2]
Causes Loss ofcross-group mixing[1]
Loses Featurecross-group mixing[1]
Presupposes Existence ofResidual Stream[1]
Lane Count8[2]
Lane Dimension96[2]
Compared toDense Linear Layer 768[2]

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.

causesLossOfblah/watt-activation/part-344
cross-group mixing
losesFeatureblah/watt-activation/part-344
cross-group mixing
partitionsResidualStreamIntoblah/watt-activation/part-344
ex:g-equals-8-independent-96-dim-lanes
presupposesExistenceOfblah/watt-activation/part-344
ex:residual-stream
labelblah/watt-activation/342
block-diagonal mode
partitionsResidualStreamIntoblah/watt-activation/342
independent lanes
laneCountblah/watt-activation/342
8
laneDimensionblah/watt-activation/342
96
comparedToblah/watt-activation/342
ex:dense-linear-layer-768

References (2)

2 references
  1. [1]Part 3444 facts
    ctx:discord/blah/watt-activation/part-344
  2. [2]3425 facts
    ctx:discord/blah/watt-activation/342
    • full textwatt-activation-342
      text/plain3 KBdoc:agent/watt-activation-342/d6e6b29a-4a18-4795-ab6d-377c23ed0f75
      Show excerpt
      [2026-03-15 23:09] xenonfun: ``` GivensRotationMixer: - Params: C(C-1)/2 angles (for C=8 groups: 28 angles per layer) - FLOPs per token: C(C-1)/2 × 4 ops (sin, cos, 2 multiply-adds) = ~112 FLOPs - Compare to dense Linear(768, 768): 5

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.