Dontopedia

Cross Patch Ar Decoder

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

Cross Patch Ar Decoder has 21 facts recorded in Dontopedia across 4 references, with 4 live disagreements.

21 facts·15 predicates·4 sources·4 in dispute

Mostly:has parameter(3), has dimension(2), has layers(2)

Maturity scale raw canonical shape-checked rule-derived certified

Inbound mentions (11)

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.

causesPerformanceIssueCauses Performance Issue(2)

decreasesOverStepsDecreases Over Steps(2)

partOfPart of(2)

concernsSubjectConcerns Subject(1)

precedesPrecedes(1)

providedPerformanceReviewOfProvided Performance Review of(1)

stableForArDecoderStable for Ar Decoder(1)

usesTeacherForcingUses Teacher Forcing(1)

Other facts (21)

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.

21 facts
PredicateValueRef
Has Parameterd=384[4]
Has Parameter3L[4]
Has Parameter15000-steps[4]
Has Dimension384[1]
Has Dimension384[2]
Has Layers3L[1]
Has Layers3[2]
Has Params3858304[1]
Has Params1305252[2]
Speed Causenow-compiled[4]
Speed Causeother-dumb-shit-removed[4]
Trained for15000 steps[1]
Trained Steps15000[2]
Context Is Training LogMlx Training[2]
Model Dimension384[3]
Layer Count3[3]
Training Steps Completed15000[3]
Parameter Count3858304[3]
Assessed on2026-03-14[3]
Has Metric1,305,252-params[4]
Has Speed Comparisonway-quicker[4]

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.

hasDimensionblah/watt-activation/part-301
384
hasLayersblah/watt-activation/part-301
3L
trainedForblah/watt-activation/part-301
15000 steps
hasParamsblah/watt-activation/part-301
3858304
trainedStepsblah/watt-activation/part-302
15000
hasLayersblah/watt-activation/part-302
3
contextIsTrainingLogblah/watt-activation/part-302
ex:mlx-training
hasDimensionblah/watt-activation/part-302
384
hasParamsblah/watt-activation/part-302
1305252
modelDimensionblah/watt-activation/299
384
layerCountblah/watt-activation/299
3
trainingStepsCompletedblah/watt-activation/299
15000
parameterCountblah/watt-activation/299
3858304
assessedOnblah/watt-activation/299
2026-03-14
hasParameterblah/watt-activation/300
d=384
hasParameterblah/watt-activation/300
3L
hasParameterblah/watt-activation/300
15000-steps
hasMetricblah/watt-activation/300
1,305,252-params
hasSpeedComparisonblah/watt-activation/300
way-quicker
speedCauseblah/watt-activation/300
now-compiled
speedCauseblah/watt-activation/300
other-dumb-shit-removed

References (4)

4 references
  1. [1]Part 3014 facts
    ctx:discord/blah/watt-activation/part-301
  2. [2]Part 3025 facts
    ctx:discord/blah/watt-activation/part-302
  3. [3]2995 facts
    ctx:discord/blah/watt-activation/299
    • full textwatt-activation-299
      text/plain2 KBdoc:agent/watt-activation-299/3c2f94b1-841d-457c-8bfa-1161d5e80363
      Show excerpt
      [2026-03-14 06:27] xenonfun: ``` Loaded 41.4 MB VQ encoder (5000 steps)... Frozen downstream: 170,329 trainable / 520,945 total step 1000 | ppl=9.0 util=0.234 | 197,267 B/s step 2000 | ppl=31.0 util=0.594 | 196,457 B/s st
  4. [4]3007 facts
    ctx:discord/blah/watt-activation/300
    • full textwatt-activation-300
      text/plain3 KBdoc:agent/watt-activation-300/3b6edccf-3524-4608-838f-25890efaea15
      Show excerpt
      [2026-03-14 06:34] xenonfun: ``` 3. Manual attention (lines 110-128) — Hand-rolled softmax attention instead of using mx.fast.scaled_dot_product_attention. MLX's fused attention kernel is significantly faster for small sequence lengths.

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.