BPB

as donto believed on

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

BPB has 88 facts recorded in Dontopedia across 56 references, with 7 live disagreements.

88 facts·73 predicates·56 sources·7 in dispute

Mostly:lower is better(6), decreases over steps(3), rdf:type(3)

Inbound mentions (36)

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.

`winsOnMetric`Wins on Metric(3)

Adamex:adam
Adamex:adam
Linearkanex:linearkan

`hasSame`Has Same(2)

Helmholtz Bandpass Fiberex:helmholtz-bandpass-fiber
V3 Lohe Syncex:v3-lohe-sync

`bpbTrackingNormally`Bpb Tracking Normally(1)

Hamex:ham

`causesImprovementIn`Causes Improvement in(1)

Seq Sizeex:seq-size

`causesRegressionIn`Causes Regression in(1)

New Custom Encodingex:new-custom-encoding

`concernsMetric`Concerns Metric(1)

Key Bpb Questionex:key-bpb-question

`demonstratesNoHarmTo`Demonstrates No Harm to(1)

Dyn H4 Phase Runex:dyn-h4-phase-run

`descendingFast`Descending Fast(1)

E8 Freshex:e8-fresh

`displaysStatCardsFor`Displays Stat Cards for(1)

Kick Runex:kick-run

`emphasizesImprovement`Emphasizes Improvement(1)

Xenonfunex:xenonfun

`emphasizesReliabilityOf`Emphasizes Reliability of(1)

Message 2026 03 20 0200ex:message-2026-03-20-0200

`exhibitsPositiveImprovement`Exhibits Positive Improvement(1)

Twtex:twt

`experiencesBriefLullOnBest`Experiences Brief Lull on Best(1)

D Muonex:d-muon

`hasMetric`Has Metric(1)

Unified Rotor Gpu 25mex:unified-rotor-gpu-25m

`hasUnitOfAdvantage`Has Unit of Advantage(1)

New Initex:new-init

`isBestOn`Is Best on(1)

Linear Kanex:linear-kan

`isConverging`Is Converging(1)

Lossex:loss

`isHealthy`Is Healthy(1)

Bpb Trajectoryex:bpb-trajectory

`maintainsSame`Maintains Same(1)

Phase Coupling Dyn H4ex:phase-coupling-dyn-h4

`matches`Matches(1)

Quality Gradientex:quality-gradient

`matchesOnMetric`Matches on Metric(1)

Hybrid Projectionex:hybrid-projection

`matchesPerformanceOn`Matches Performance on(1)

Hybrid Projectionex:hybrid-projection

`measures`Measures(1)

Run Oneex:run-one

`measuresMetric`Measures Metric(1)

Required Run 1ex:required-run-1

`monitors`Monitors(1)

Run Twoex:run-two

`monitorsMetric`Monitors Metric(1)

Required Run 2ex:required-run-2

`ordersBy`Orders by(1)

Current Session Leaderboardex:current-session-leaderboard

`providesStandardMetric`Provides Standard Metric(1)

Salon Validationex:salon-validation

`removingFixedCouplingDoesNotHurt`Removing Fixed Coupling Does Not Hurt(1)

Dyn H4 Phase Runex:dyn-h4-phase-run

`shouldImprove`Should Improve(1)

Increasing Seqex:increasing-seq

`showsTieOn`Shows Tie on(1)

Step 950 Metricsex:step-950-metrics

`tradesOffAgainst`Trades Off Against(1)

Throughputex:throughput

`willProbablyRefine`Will Probably Refine(1)

E7 Resumeex:e7-resume

Other facts (85)

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.

85 facts

Predicate	Value	Ref
Lower Is Better	null	[2]
Lower Is Better	true	[3]
Lower Is Better	true	[8]
Lower Is Better	True	[29]
Lower Is Better	null	[30]
Lower Is Better	true	[48]
Decreases Over Steps	Cross Patch Ar Decoder	[1]
Decreases Over Steps	Step 1000 to 5000	[9]
Decreases Over Steps	null	[15]
Rdf:type	Performance Metric	[52]
Rdf:type	Metric	[54]
Rdf:type	Metric	[55]
Improved to	4.43	[6]
Improved to	1.2146	[43]
Primary Metric	null	[27]
Primary Metric	Training Evaluation	[35]
Measures Compression	bits per byte	[45]
Measures Compression	Language Model	[47]
Measures Bits Per Byte	{}	[1]
Uses Log Base	2	[4]
Is More Intuitive Than Nats Per Byte	null	[4]
Random Data Value	8	[4]
Theoretical Minimum	0	[4]
Defined As	CE_loss / ln(2)	[4]
Equals Nats Per Byte Divided by	ln(2)	[4]
Interpretable As Bits to Encode Each Byte	null	[4]
Value at Step200	2.99	[5]
Improving Rapidly	true	[5]
Decreased Over Steps	1100 Steps	[6]
Improved From	4.86	[6]
Means Bits Per Byte	True	[7]
Improves With Depth	true	[10]
Is Valid Metric	true	[11]
References Performance Metric	Bits Per Byte	[12]
Stable in Ham	Ham	[13]
Shows V3 Wins by	0.07	[14]
Has Value for L2	5.20	[16]
Drops Steadily in Gated	Gated Rotor	[17]
Value at Step3700	3.67	[18]
Correlates Inversely With Quality	{}	[19]
Is Key Quality Metric	{}	[19]
Improves Over	Old Init	[20]
Drops Steadily From to	[5.7,5.34]	[20]
Changes From to in Steps	[7.12,5.34,275]	[20]
Value at Step3400	3.33	[21]
Improving	Current Run	[21]
Presupposes Acronym	Bits Per Byte	[22]
Measures	Language Modeling Performance	[22]
Improves Later Than	Snr Capacity	[23]
Primary Metric for	Run Comparison	[24]
Used for Comparison Despite Naming Differences	true	[24]
Is Same	V3 and Helmholtz	[25]
Measured in Bits Per Byte	true	[26]
Is Noisy	true	[26]
Bounces Between	5.0 and 8.0	[26]
Decreased From	6.44	[28]
Measures Performance	Model	[28]
Implies Compression Quality	Model Performance	[28]
Decreased From Earlier	6.61	[28]
Improved	true	[28]
Improved by	0.8	[28]
Is Evaluation Metric	Bits Per Byte	[31]
Is Primary Metric	null	[32]
Exists As Performance Metric	Language Model Training	[33]
Measures Quality	true	[34]
Lacks Separation Capability	Fluent Vs Degenerate	[36]
Does Not Separate Fluent From Degenerate	alone	[36]
Refers to Bits Per Byte	null	[37]
Measures Compression Efficiency	Yes	[38]
Primary Benchmark	Model Evaluation	[39]
Standard in Language Model Eval	true	[40]
Improved to Best	1.0219	[41]
Remains Sane	true	[42]
Is Noisy Throughout	true	[42]
Is Bits Per Byte Metric	null	[44]
Decreases Monotonically Almost	Steps 1 to 750	[46]
Might Plateau at	2.4	[49]
Might Break Below	2.3	[49]
Is Performance Metric	null	[50]
Is Bits Per Byte	true	[51]
Abbreviation for	Bits Per Byte	[53]
Drops Monotonically	true	[54]
Start Value	3.00	[54]
End Value	2.53	[54]
Value Change Range	D=6→D=32	[54]

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.

88 undated

—

decreasesOverStepsblah/watt-activation/part-301

ex:cross-patch-ar-decoder

—

measuresBitsPerByteblah/watt-activation/part-301

{}

—

lowerIsBetterblah/watt-activation/part-303

null

—

lowerIsBetterblah/watt-activation/part-323

true

—

usesLogBaseblah/watt-activation/part-327

—

isMoreIntuitiveThanNatsPerByteblah/watt-activation/part-327

null

—

randomDataValueblah/watt-activation/part-327

—

theoreticalMinimumblah/watt-activation/part-327

—

definedAsblah/watt-activation/part-327

CE_loss / ln(2)

—

equalsNatsPerByteDividedByblah/watt-activation/part-327

ln(2)

—

interpretableAsBitsToEncodeEachByteblah/watt-activation/part-327

null

—

valueAtStep200blah/watt-activation/part-322

2.99

—

improvingRapidlyblah/watt-activation/part-322

true

—

improvedToblah/watt-activation/part-338

4.43

—

decreasedOverStepsblah/watt-activation/part-338

ex:1100-steps

—

improvedFromblah/watt-activation/part-338

4.86

—

meansBitsPerByteblah/watt-activation/part-343

ex:true

—

lowerIsBetterblah/watt-activation/part-345

true

—

decreasesOverStepsblah/watt-activation/part-355

ex:step-1000-to-5000

—

improvesWithDepthblah/watt-activation/part-357

true

—

isValidMetricblah/watt-activation/part-360

true

—

referencesPerformanceMetricblah/watt-activation/part-351

ex:bits-per-byte

—

stableInHamblah/watt-activation/part-376

ex:ham

—

showsV3WinsByblah/watt-activation/part-383

0.07

—

decreasesOverStepsblah/watt-activation/part-377

null

—

hasValueForL2blah/watt-activation/part-391

5.20

—

dropsSteadilyInGatedblah/watt-activation/part-400

ex:gated-rotor

—

valueAtStep3700blah/watt-activation/part-397

3.67

—

correlatesInverselyWithQualityblah/watt-activation/part-407

{}

—

isKeyQualityMetricblah/watt-activation/part-407

{}

—

improvesOverblah/watt-activation/part-418

ex:old-init

—

dropsSteadilyFromToblah/watt-activation/part-418

[5.7,5.34]

—

changesFromToInStepsblah/watt-activation/part-418

[7.12,5.34,275]

—

valueAtStep3400blah/watt-activation/part-413

3.33

—

improvingblah/watt-activation/part-413

ex:current-run

—

presupposesAcronymblah/watt-activation/part-427

ex:bits-per-byte

—

measuresblah/watt-activation/part-427

ex:language-modeling-performance

—

improvesLaterThanblah/watt-activation/part-425

ex:snr-capacity

—

primaryMetricForblah/watt-activation/part-428

ex:run-comparison

—

usedForComparisonDespiteNamingDifferencesblah/watt-activation/part-428

true

—

isSameblah/watt-activation/part-453

ex:v3-and-helmholtz

—

measuredInBitsPerByteblah/watt-activation/part-462

true

—

isNoisyblah/watt-activation/part-462

true

—

bouncesBetweenblah/watt-activation/part-462

5.0 and 8.0

—

primaryMetricblah/watt-activation/part-472

null

—

decreasedFromblah/watt-activation/part-494

6.44

—

measuresPerformanceblah/watt-activation/part-494

ex:model

—

impliesCompressionQualityblah/watt-activation/part-494

ex:model-performance

—

decreasedFromEarlierblah/watt-activation/part-494

6.61

—

improvedblah/watt-activation/part-494

true

—

improvedByblah/watt-activation/part-494

0.8

—

lowerIsBetterblah/watt-activation/part-496

ex:true

—

lowerIsBetterblah/watt-activation/part-609

null

—

isEvaluationMetricblah/watt-activation/part-634

ex:bits-per-byte

—

isPrimaryMetricblah/watt-activation/part-648

null

—

existsAsPerformanceMetricblah/watt-activation/part-644

ex:language-model-training

—

measuresQualityblah/watt-activation/part-664

true

—

primaryMetricblah/watt-activation/part-662

ex:training-evaluation

—

lacksSeparationCapabilityblah/watt-activation/part-675

ex:fluent-vs-degenerate

—

doesNotSeparateFluentFromDegenerateblah/watt-activation/part-675

alone

—

refersToBitsPerByteblah/watt-activation/part-670

null

—

measuresCompressionEfficiencyblah/watt-activation/part-689

ex:yes

—

primaryBenchmarkblah/watt-activation/part-696

ex:model-evaluation

—

standardInLanguageModelEvalblah/watt-activation/part-686

true

—

improvedToBestblah/watt-activation/part-703

1.0219

—

remainsSaneblah/watt-activation/part-705

true

—

isNoisyThroughoutblah/watt-activation/part-705

true

—

improvedToblah/watt-activation/part-707

1.2146

—

isBitsPerByteMetricblah/watt-activation/part-710

null

—

measuresCompressionblah/watt-activation/part-713

bits per byte

—

decreasesMonotonicallyAlmostblah/watt-activation/part-700

ex:steps-1-to-750

—

measuresCompressionblah/watt-activation/part-302

ex:language-model

—

lowerIsBetterblah/watt-activation/part-367

true

—

mightPlateauAtblah/watt-activation/part-325

2.4

—

mightBreakBelowblah/watt-activation/part-325

2.3

—

isPerformanceMetricblah/watt-activation/part-475

null

—

isBitsPerByteblah/watt-activation/part-643

true

—

typeblah/watt-activation/323

ex:PerformanceMetric

—

abbreviationForblah/watt-activation/326

ex:bits-per-byte

—

labelblah/watt-activation/355

BPB

—

typeblah/watt-activation/355

ex:Metric

—

dropsMonotonicallyblah/watt-activation/355

true

—

startValueblah/watt-activation/355

3.00

—

endValueblah/watt-activation/355

2.53

—

valueChangeRangeblah/watt-activation/355

D=6→D=32

—

typeblah/watt-activation/358

ex:Metric

—

labelblah/watt-activation/358

BPB

—

labelblah/watt-activation/679

Bits Per Byte

References (56)

56 references

[1]Part 3012 facts
ctx:discord/blah/watt-activation/part-301
[2]Part 3031 fact
ctx:discord/blah/watt-activation/part-303
[3]Part 3231 fact
ctx:discord/blah/watt-activation/part-323
[4]Part 3277 facts
ctx:discord/blah/watt-activation/part-327
[5]Part 3222 facts
ctx:discord/blah/watt-activation/part-322
[6]Part 3383 facts
ctx:discord/blah/watt-activation/part-338
[7]Part 3431 fact
ctx:discord/blah/watt-activation/part-343
[8]Part 3451 fact
ctx:discord/blah/watt-activation/part-345
[9]Part 3551 fact
ctx:discord/blah/watt-activation/part-355
[10]Part 3571 fact
ctx:discord/blah/watt-activation/part-357
[11]Part 3601 fact
ctx:discord/blah/watt-activation/part-360
[12]Part 3511 fact
ctx:discord/blah/watt-activation/part-351
[13]Part 3761 fact
ctx:discord/blah/watt-activation/part-376
[14]Part 3831 fact
ctx:discord/blah/watt-activation/part-383
[15]Part 3771 fact
ctx:discord/blah/watt-activation/part-377
[16]Part 3911 fact
ctx:discord/blah/watt-activation/part-391
[17]Part 4001 fact
ctx:discord/blah/watt-activation/part-400
[18]Part 3971 fact
ctx:discord/blah/watt-activation/part-397
[19]Part 4072 facts
ctx:discord/blah/watt-activation/part-407
[20]Part 4183 facts
ctx:discord/blah/watt-activation/part-418
[21]Part 4132 facts
ctx:discord/blah/watt-activation/part-413
[22]Part 4272 facts
ctx:discord/blah/watt-activation/part-427
[23]Part 4251 fact
ctx:discord/blah/watt-activation/part-425
[24]Part 4282 facts
ctx:discord/blah/watt-activation/part-428
[25]Part 4531 fact
ctx:discord/blah/watt-activation/part-453
[26]Part 4623 facts
ctx:discord/blah/watt-activation/part-462
[27]Part 4721 fact
ctx:discord/blah/watt-activation/part-472
[28]Part 4946 facts
ctx:discord/blah/watt-activation/part-494
[29]Part 4961 fact
ctx:discord/blah/watt-activation/part-496
[30]Part 6091 fact
ctx:discord/blah/watt-activation/part-609
[31]Part 6341 fact
ctx:discord/blah/watt-activation/part-634
[32]Part 6481 fact
ctx:discord/blah/watt-activation/part-648
[33]Part 6441 fact
ctx:discord/blah/watt-activation/part-644
[34]Part 6641 fact
ctx:discord/blah/watt-activation/part-664
[35]Part 6621 fact
ctx:discord/blah/watt-activation/part-662
[36]Part 6752 facts
ctx:discord/blah/watt-activation/part-675
[37]Part 6701 fact
ctx:discord/blah/watt-activation/part-670
[38]Part 6891 fact
ctx:discord/blah/watt-activation/part-689
[39]Part 6961 fact
ctx:discord/blah/watt-activation/part-696
[40]Part 6861 fact
ctx:discord/blah/watt-activation/part-686
[41]Part 7031 fact
ctx:discord/blah/watt-activation/part-703
[42]Part 7052 facts
ctx:discord/blah/watt-activation/part-705
[43]Part 7071 fact
ctx:discord/blah/watt-activation/part-707
[44]Part 7101 fact
ctx:discord/blah/watt-activation/part-710
[45]Part 7131 fact
ctx:discord/blah/watt-activation/part-713
[46]Part 7001 fact
ctx:discord/blah/watt-activation/part-700
[47]Part 3021 fact
ctx:discord/blah/watt-activation/part-302
[48]Part 3671 fact
ctx:discord/blah/watt-activation/part-367
[49]Part 3252 facts
ctx:discord/blah/watt-activation/part-325
[50]Part 4751 fact
ctx:discord/blah/watt-activation/part-475
[51]Part 6431 fact
ctx:discord/blah/watt-activation/part-643
[52]3231 fact
ctx:discord/blah/watt-activation/323
- full textwatt-activation-323
  text/plain3 KBdoc:agent/watt-activation-323/60722a8e-0dbe-4aee-be7b-862e43a25b6f
  Show excerpt
  [2026-03-15 03:41] xenonfun: ``` Elevator Pitch We built a language model that doesn't use a tokenizer. Instead of BPE or word pieces, we encode raw bytes directly as phase-modulated oscillators using digital communications theory…
[53]3261 fact
ctx:discord/blah/watt-activation/326
- full textwatt-activation-326
  text/plain3 KBdoc:agent/watt-activation-326/13bd929c-8a9b-493f-b61c-5db579a2f719
  Show excerpt
  [2026-03-15 04:22] xenonfun: # What The Literature Review Means For This Result The timing is significant. The search found that: - AKOrN (ICLR 2025) validated Kuramoto oscillators in transformers - SSA (February 2026) showed Kuramoto atte…
[54]3556 facts
ctx:discord/blah/watt-activation/355
- full textwatt-activation-355
  text/plain3 KBdoc:agent/watt-activation-355/e62c81a8-1082-4c07-b675-3759a8600d0e
  Show excerpt
  [2026-03-17 15:27] xenonfun: ``` Key findings: 1. Depth scaling is smooth and strong: BPB drops monotonically 3.00→2.53 from D=6→D=32. DC@16 rises 72%→91%. 2. Retrieval reach = 128 for ALL configs — every model retrieves across the f…
[55]3582 facts
ctx:discord/blah/watt-activation/358
- full textwatt-activation-358
  text/plain2 KBdoc:agent/watt-activation-358/5e6372ad-404d-41bf-87ea-6453f5db1dc5
  Show excerpt
  [2026-03-17 17:23] xenonfun: This is a decisive result for the antenna branch. Updated conclusion: - For the antenna architecture in the tested regime, RotAdamW / strict sphere ∩ zero-mean manifold constraint is actively harmful. - Adam wi…
[56]6791 fact
ctx:discord/blah/watt-activation/679
- full textwatt-activation-679
  text/plain2 KBdoc:agent/watt-activation-679/e7b96ce5-bed6-4b84-baf1-a773f080c0db
  Show excerpt
  [2026-04-22 14:42] xenonfun: ``` ⏺ Pulling the per-step bpb trajectory from metrics.jsonl of the two most-comparable prior runs: step | unified_rotor_gpu_25m | lohe_delta_25m_replication | lohe_delta_16L_G16_H4_R16_doremi …

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.

BPB

Inbound mentions (36)

winsOnMetricWins on Metric(3)

hasSameHas Same(2)

bpbTrackingNormallyBpb Tracking Normally(1)

causesImprovementInCauses Improvement in(1)

causesRegressionInCauses Regression in(1)

concernsMetricConcerns Metric(1)

demonstratesNoHarmToDemonstrates No Harm to(1)

descendingFastDescending Fast(1)

displaysStatCardsForDisplays Stat Cards for(1)

emphasizesImprovementEmphasizes Improvement(1)

emphasizesReliabilityOfEmphasizes Reliability of(1)

exhibitsPositiveImprovementExhibits Positive Improvement(1)

experiencesBriefLullOnBestExperiences Brief Lull on Best(1)

hasMetricHas Metric(1)

hasUnitOfAdvantageHas Unit of Advantage(1)

isBestOnIs Best on(1)

isConvergingIs Converging(1)

isHealthyIs Healthy(1)

maintainsSameMaintains Same(1)

matchesMatches(1)

matchesOnMetricMatches on Metric(1)

matchesPerformanceOnMatches Performance on(1)

measuresMeasures(1)

measuresMetricMeasures Metric(1)

monitorsMonitors(1)

monitorsMetricMonitors Metric(1)

ordersByOrders by(1)

providesStandardMetricProvides Standard Metric(1)

removingFixedCouplingDoesNotHurtRemoving Fixed Coupling Does Not Hurt(1)

shouldImproveShould Improve(1)

showsTieOnShows Tie on(1)

tradesOffAgainstTrades Off Against(1)

willProbablyRefineWill Probably Refine(1)

Other facts (85)

Timeline

References (56)

See also

Keep researching

`winsOnMetric`Wins on Metric(3)

`hasSame`Has Same(2)

`bpbTrackingNormally`Bpb Tracking Normally(1)

`causesImprovementIn`Causes Improvement in(1)

`causesRegressionIn`Causes Regression in(1)

`concernsMetric`Concerns Metric(1)

`demonstratesNoHarmTo`Demonstrates No Harm to(1)

`descendingFast`Descending Fast(1)

`displaysStatCardsFor`Displays Stat Cards for(1)

`emphasizesImprovement`Emphasizes Improvement(1)

`emphasizesReliabilityOf`Emphasizes Reliability of(1)

`exhibitsPositiveImprovement`Exhibits Positive Improvement(1)

`experiencesBriefLullOnBest`Experiences Brief Lull on Best(1)

`hasMetric`Has Metric(1)

`hasUnitOfAdvantage`Has Unit of Advantage(1)

`isBestOn`Is Best on(1)

`isConverging`Is Converging(1)

`isHealthy`Is Healthy(1)

`maintainsSame`Maintains Same(1)

`matches`Matches(1)

`matchesOnMetric`Matches on Metric(1)

`matchesPerformanceOn`Matches Performance on(1)

`measures`Measures(1)

`measuresMetric`Measures Metric(1)

`monitors`Monitors(1)

`monitorsMetric`Monitors Metric(1)

`ordersBy`Orders by(1)

`providesStandardMetric`Provides Standard Metric(1)

`removingFixedCouplingDoesNotHurt`Removing Fixed Coupling Does Not Hurt(1)

`shouldImprove`Should Improve(1)

`showsTieOn`Shows Tie on(1)

`tradesOffAgainst`Trades Off Against(1)

`willProbablyRefine`Will Probably Refine(1)