Dontopedia

Ppl

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

Ppl has 57 facts recorded in Dontopedia across 49 references, with 6 live disagreements.

57 facts·48 predicates·49 sources·6 in dispute

Mostly:decreases over steps(4), decreases over iterations(3), correlates with loss(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.

assumesKnowledgeOfAssumes Knowledge of(1)

decreasesOverStepsDecreases Over Steps(1)

hasLowerHas Lower(1)

improvesPerformanceImproves Performance(1)

measuresMetricMeasures Metric(1)

movesDownwardInMoves Downward in(1)

nowHasMetricNow Has Metric(1)

plansFurtherTestingPlans Further Testing(1)

showsDecreasingPerplexityShows Decreasing Perplexity(1)

showsGradualImprovementShows Gradual Improvement(1)

tracksMetricTracks Metric(1)

Other facts (57)

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.

57 facts
PredicateValueRef
Decreases Over StepsSteps 1000 to 2000[27]
Decreases Over StepsStep 100 to 500[32]
Decreases Over Steps151k to 2.5k[34]
Decreases Over Stepsnull[37]
Decreases Over IterationsE Mhkan H5 Training Run[17]
Decreases Over IterationsIter 40500 to 42000[25]
Decreases Over IterationsTrue[47]
Correlates With LossTraining Run[3]
Correlates With LossTrue[33]
Derives From LossAvg Loss[6]
Derives From LossLoss[45]
Decreases Over ItersIter 500 to 2000[22]
Decreases Over ItersAnchor V3 M32 L2048[46]
References Perplexity Loss MetricML Evaluation[23]
References Perplexity Loss MetricPpl Metric[23]
Means Perplexitynull[1]
Is43 8 at Final Val43.8[2]
Derived From Losstrue[4]
Exceeds Value1200[5]
Is Competitive~72[7]
Used As Performance Metricnull[8]
Decreases During Training↓[9]
Lower Is Better{}[10]
Training MetricLong Seq[11]
Is Worse When Highernull[12]
Is Key MetricTrue[13]
Measures Performancenull[14]
Is Lower inAnchor[15]
Lower Value Is Bettertrue[16]
Is Performance MetricLower Better[16]
Primary Metricnull[18]
Expected UnchangedNew Formulation[19]
Remains Unchanged Across Implsnull[20]
Was MeasuredAfter Changes[21]
Remains UnchangedAfter Optimizations[21]
Is IdenticalOld Implementation[23]
Used As Quality MetricIdentical to Old[23]
Drops Steadilytrue[24]
Is Identical Indicating EquivalenceOld Implementation[23]
Serves As Progress Metricnull[26]
Became Nan AfterTraining Step 10300[28]
Increases FromStep 10100[29]
Indicates Weak Performancetrue[30]
Diverges to35M[31]
Correlates Inversely With Lossnull[35]
Indicates High Perplexity83.4[36]
Measures Model PerformanceTraining Session[36]
Is Final Metricnull[38]
Perplexity Metrictrue[39]
Measures Perplexity{}[40]
Increases From Step1000 To2000Vq Encoder[40]
Is Quality Metricnull[41]
Is MetricWire Encoding Results[42]
Current Value4.55[43]
Decreased Over Steps131 to 44[44]
Is Learning Curve Metricnull[48]
Expanded FormPerplexity[49]

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.

57 undated
—
meansPerplexityblah/random/part-27
null
—
is43-8AtFinalValblah/safiersemantics/part-72
43.8
—
correlatesWithLossblah/watt-activation/part-21
ex:training-run
—
derivedFromLossblah/watt-activation/part-25
true
—
exceedsValueblah/watt-activation/part-44
1200
—
derivesFromLossblah/watt-activation/part-47
ex:avg-loss
—
isCompetitiveblah/watt-activation/part-48
~72
—
usedAsPerformanceMetricblah/watt-activation/part-50
null
—
decreasesDuringTrainingblah/watt-activation/part-55
↓
—
lowerIsBetterblah/watt-activation/part-52
{}
—
trainingMetricblah/watt-activation/part-57
ex:long-seq
—
isWorseWhenHigherblah/watt-activation/part-46
null
—
isKeyMetricblah/watt-activation/part-60
ex:true
—
measuresPerformanceblah/watt-activation/part-65
null
—
isLowerInblah/watt-activation/part-54
ex:anchor
—
lowerValueIsBetterblah/watt-activation/part-68
true
—
isPerformanceMetricblah/watt-activation/part-68
ex:lower-better
—
decreasesOverIterationsblah/watt-activation/part-71
ex:e-mhkan-h5-training-run
—
primaryMetricblah/watt-activation/part-69
null
—
expectedUnchangedblah/watt-activation/part-74
ex:new-formulation
—
remainsUnchangedAcrossImplsblah/watt-activation/part-76
null
—
wasMeasuredblah/watt-activation/part-64
ex:after-changes
—
remainsUnchangedblah/watt-activation/part-64
ex:after-optimizations
—
decreasesOverItersblah/watt-activation/part-84
ex:iter-500-to-2000
—
isIdenticalblah/watt-activation/part-77
ex:old-implementation
—
usedAsQualityMetricblah/watt-activation/part-77
ex:identical-to-old
—
referencesPerplexityLossMetricblah/watt-activation/part-77
ex:ml-evaluation
—
dropsSteadilyblah/watt-activation/part-94
true
—
decreasesOverIterationsblah/watt-activation/part-98
ex:iter-40500-to-42000
—
referencesPerplexityLossMetricblah/watt-activation/part-77
ex:ppl-metric
—
isIdenticalIndicatingEquivalenceblah/watt-activation/part-77
ex:old-implementation
—
servesAsProgressMetricblah/watt-activation/part-95
null
—
decreasesOverStepsblah/watt-activation/part-126
ex:steps-1000-to-2000
—
becameNanAfterblah/watt-activation/part-136
ex:training-step-10300
—
increasesFromblah/watt-activation/part-137
ex:step-10100
—
indicatesWeakPerformanceblah/watt-activation/part-172
true
—
divergesToblah/watt-activation/part-194
35M
—
decreasesOverStepsblah/watt-activation/part-189
ex:step-100-to-500
—
correlatesWithLossblah/watt-activation/part-210
ex:true
—
decreasesOverStepsblah/watt-activation/part-202
151k to 2.5k
—
correlatesInverselyWithLossblah/watt-activation/part-217
null
—
indicatesHighPerplexityblah/watt-activation/part-245
83.4
—
measuresModelPerformanceblah/watt-activation/part-245
ex:training-session
—
decreasesOverStepsblah/watt-activation/part-266
null
—
isFinalMetricblah/watt-activation/part-279
null
—
perplexityMetricblah/watt-activation/part-289
true
—
measuresPerplexityblah/watt-activation/part-301
{}
—
increasesFromStep1000To2000blah/watt-activation/part-301
ex:vq-encoder
—
isQualityMetricblah/watt-activation/part-317
null
—
isMetricblah/watt-activation/part-321
ex:wire-encoding-results
—
currentValueblah/watt-activation/part-397
4.55
—
decreasedOverStepsblah/training-and-evals/part-38
131 to 44
—
derivesFromLossblah/watt-activation/part-37
ex:loss
—
decreasesOverItersblah/watt-activation/part-61
ex:anchor-v3-m32-l2048
—
decreasesOverIterationsblah/watt-activation/part-86
ex:true
—
isLearningCurveMetricblah/watt-activation/part-319
null
—
expandedFormblah/watt-activation/77
ex:perplexity

References (49)

49 references
  1. [1]Part 271 fact
    ctx:discord/blah/random/part-27
  2. [2]Part 721 fact
    ctx:discord/blah/safiersemantics/part-72
  3. [3]Part 211 fact
    ctx:discord/blah/watt-activation/part-21
  4. [4]Part 251 fact
    ctx:discord/blah/watt-activation/part-25
  5. [5]Part 441 fact
    ctx:discord/blah/watt-activation/part-44
  6. [6]Part 471 fact
    ctx:discord/blah/watt-activation/part-47
  7. [7]Part 481 fact
    ctx:discord/blah/watt-activation/part-48
  8. [8]Part 501 fact
    ctx:discord/blah/watt-activation/part-50
  9. [9]Part 551 fact
    ctx:discord/blah/watt-activation/part-55
  10. [10]Part 521 fact
    ctx:discord/blah/watt-activation/part-52
  11. [11]Part 571 fact
    ctx:discord/blah/watt-activation/part-57
  12. [12]Part 461 fact
    ctx:discord/blah/watt-activation/part-46
  13. [13]Part 601 fact
    ctx:discord/blah/watt-activation/part-60
  14. [14]Part 651 fact
    ctx:discord/blah/watt-activation/part-65
  15. [15]Part 541 fact
    ctx:discord/blah/watt-activation/part-54
  16. [16]Part 682 facts
    ctx:discord/blah/watt-activation/part-68
  17. [17]Part 711 fact
    ctx:discord/blah/watt-activation/part-71
  18. [18]Part 691 fact
    ctx:discord/blah/watt-activation/part-69
  19. [19]Part 741 fact
    ctx:discord/blah/watt-activation/part-74
  20. [20]Part 761 fact
    ctx:discord/blah/watt-activation/part-76
  21. [21]Part 642 facts
    ctx:discord/blah/watt-activation/part-64
  22. [22]Part 841 fact
    ctx:discord/blah/watt-activation/part-84
  23. [23]Part 775 facts
    ctx:discord/blah/watt-activation/part-77
  24. [24]Part 941 fact
    ctx:discord/blah/watt-activation/part-94
  25. [25]Part 981 fact
    ctx:discord/blah/watt-activation/part-98
  26. [26]Part 951 fact
    ctx:discord/blah/watt-activation/part-95
  27. [27]Part 1261 fact
    ctx:discord/blah/watt-activation/part-126
  28. [28]Part 1361 fact
    ctx:discord/blah/watt-activation/part-136
  29. [29]Part 1371 fact
    ctx:discord/blah/watt-activation/part-137
  30. [30]Part 1721 fact
    ctx:discord/blah/watt-activation/part-172
  31. [31]Part 1941 fact
    ctx:discord/blah/watt-activation/part-194
  32. [32]Part 1891 fact
    ctx:discord/blah/watt-activation/part-189
  33. [33]Part 2101 fact
    ctx:discord/blah/watt-activation/part-210
  34. [34]Part 2021 fact
    ctx:discord/blah/watt-activation/part-202
  35. [35]Part 2171 fact
    ctx:discord/blah/watt-activation/part-217
  36. [36]Part 2452 facts
    ctx:discord/blah/watt-activation/part-245
  37. [37]Part 2661 fact
    ctx:discord/blah/watt-activation/part-266
  38. [38]Part 2791 fact
    ctx:discord/blah/watt-activation/part-279
  39. [39]Part 2891 fact
    ctx:discord/blah/watt-activation/part-289
  40. [40]Part 3012 facts
    ctx:discord/blah/watt-activation/part-301
  41. [41]Part 3171 fact
    ctx:discord/blah/watt-activation/part-317
  42. [42]Part 3211 fact
    ctx:discord/blah/watt-activation/part-321
  43. [43]Part 3971 fact
    ctx:discord/blah/watt-activation/part-397
  44. [44]Part 381 fact
    ctx:discord/blah/training-and-evals/part-38
  45. [45]Part 371 fact
    ctx:discord/blah/watt-activation/part-37
  46. [46]Part 611 fact
    ctx:discord/blah/watt-activation/part-61
  47. [47]Part 861 fact
    ctx:discord/blah/watt-activation/part-86
  48. [48]Part 3191 fact
    ctx:discord/blah/watt-activation/part-319
  49. [49]771 fact
    ctx:discord/blah/watt-activation/77
    • full textwatt-activation-77
      text/plain3 KBdoc:agent/watt-activation-77/59dddcca-2e06-4d74-9254-03846e959489
      Show excerpt
      [2026-03-07 18:47] xenonfun: It wrote a fused metal kernel, tho I think we are going to have it also do mlx path shortly: ``` Architecture Summary Training path — chunked prefix scan: - _anchor_kan_forward_chunked() processes chunks o…

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.

↑