DoReMi

as donto believed on

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

DoReMi has 78 facts recorded in Dontopedia across 24 references, with 5 live disagreements.

78 facts·44 predicates·24 sources·5 in dispute

Mostly:has training step(18), uses dataset(6), processes dataset(6)

Has Training Stepin dispute`hasTrainingStep`

Training Step 1900^[17]all time · Part 711
Step 1750^[17]all time · Part 711
Training Step 1760^[17]all time · Part 711
Training Step 1770^[17]all time · Part 711
Training Step 1780^[17]all time · Part 711
Training Step 1790^[17]all time · Part 711
Training Step 1800^[17]all time · Part 711
Training Step 1810^[17]all time · Part 711
Training Step 1820^[17]all time · Part 711
Training Step 1830^[17]all time · Part 711

Inbound mentions (15)

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.

`partOfRun`Part of Run(2)

Training Step 1750ex:training-step-1750
Training Step 1760ex:training-step-1760

`basedOn`Based on(1)

Teacher in Loop Agentex:teacher-in-loop-agent

`capsDoReMiByDomainSize`Caps Do Re Mi by Domain Size(1)

Direction 1ex:direction-1

`combines`Combines(1)

Item 4ex:item-4

`concerns`Concerns(1)

Log Entryex:log-entry

`involvesFeature`Involves Feature(1)

Item 4 Orderex:item-4-order

`letsDoReMiReweight`Lets Do Re Mi Reweight(1)

Direction 1ex:direction-1

`mentionsSubject`Mentions Subject(1)

Log Entry 2026 05 02 01 01ex:log-entry-2026-05-02-01-01

`reportsProgressOf`Reports Progress of(1)

Log Entryex:log-entry

`restoredCleanly`Restored Cleanly(1)

Resumed Runex:resumed-run

`restoredState`Restored State(1)

Resumed Runex:resumed-run

`signalsHigherLoss`Signals Higher Loss(1)

↑ Rankingex:↑-ranking

`usesMethod`Uses Method(1)

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

`worksAsDesigned`Works As Designed(1)

Unified Rotor Gpu 25m Doremiex:unified-rotor-gpu-25m-doremi

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

Predicate	Value	Ref
Uses Dataset	Classics Literature	[17]
Uses Dataset	General Knowledge	[17]
Uses Dataset	Practical Texts	[17]
Uses Dataset	Professional Benchmark	[17]
Uses Dataset	Reference Web	[17]
Uses Dataset	Speculative Fiction	[17]
Processes Dataset	General Knowledge	[24]
Processes Dataset	Reference Web	[24]
Processes Dataset	Classics Literature	[24]
Processes Dataset	Professional Benchmark	[24]
Processes Dataset	Practical Texts	[24]
Processes Dataset	Speculative Fiction	[24]
Rdf:type	Software Component	[20]
Rdf:type	Training Module	[21]
Rdf:type	Algorithm	[22]
Rdf:type	Weighting Algorithm	[23]
Rdf:type	Model Run	[24]
Is Known Technique	Curriculum Training	[1]
Included in	Item 4	[2]
Is Known Method	true	[3]
Is Weight Update Method	Domain Adaptation	[4]
References Optimization Technique	Doremi Weights	[5]
Bets on Domain Sensitivity	null	[6]
Referenced in Context	null	[6]
Reweighted After Step	1000	[7]
Has Designed Functionality	Dataset Weighting	[8]
Heavily Weighted on Hard Domains	true	[9]
Up Weights Harder Ones	true	[10]
Creates Feedback Loop	true	[10]
Presupposed As Training Method	true	[10]
Deontically Required for Hard Domain Handling	true	[10]
Sees More Per Domain Loss	true	[10]
Restorable Cleanly	True	[11]
Causes Tiny Domain Overexposure	Tiny Domains	[12]
Adaptive Weighting	true	[12]
Reweights Domains	Domains	[12]
Is Integrated Component	null	[13]
Indicates Trend Up	Poetry Dataset	[14]
Indicates Trend Down	Simple Educational Dataset	[14]
Reacts to	Overexposure	[15]
Prioritizes Harder Data	true	[15]
Has Started Lowering	Professional Benchmark	[15]
Correctly Adjusts Weights	true	[15]
References Domain Reweighting Method	prior technique	[16]
Presupposes Existence	true	[16]
Computes	Raw Domain Weights	[16]
Maintains Constant Learning Rate	0.34	[17]
Can Upweight	Hard Domains	[18]
Handles Domain Weighting	True	[18]
Is Sampling Method	True	[18]
Prediction Hypothesis	domain-sensitive AND fluent	[19]
Can Upweight Hard Domains	true	[22]
Capable of	noticing hard/noisy tiny domains	[23]
Limited by	Exposure Governor	[23]
Has Training Start Step	1750	[24]
Has Training End Step	1920	[24]
Uses Exposure Governor	Sampling Weights After Exposure Governor	[24]

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.

78 undated

—

isKnownTechniqueblah/watt-activation/part-634

ex:curriculum-training

—

includedInblah/watt-activation/part-640

ex:item-4

—

isKnownMethodblah/watt-activation/part-645

true

—

isWeightUpdateMethodblah/watt-activation/part-662

ex:domain-adaptation

—

referencesOptimizationTechniqueblah/watt-activation/part-668

ex:doremi-weights

—

betsOnDomainSensitivityblah/watt-activation/part-670

null

—

referencedInContextblah/watt-activation/part-670

null

—

reweightedAfterStepblah/watt-activation/part-680

1000

—

hasDesignedFunctionalityblah/watt-activation/part-682

ex:dataset-weighting

—

heavilyWeightedOnHardDomainsblah/watt-activation/part-687

true

—

upWeightsHarderOnesblah/watt-activation/part-686

true

—

createsFeedbackLoopblah/watt-activation/part-686

true

—

presupposedAsTrainingMethodblah/watt-activation/part-686

true

—

deonticallyRequiredForHardDomainHandlingblah/watt-activation/part-686

true

—

seesMorePerDomainLossblah/watt-activation/part-686

true

—

restorableCleanlyblah/watt-activation/part-702

ex:true

—

causesTinyDomainOverexposureblah/watt-activation/part-703

ex:tiny-domains

—

adaptiveWeightingblah/watt-activation/part-703

true

—

reweightsDomainsblah/watt-activation/part-703

ex:domains

—

isIntegratedComponentblah/watt-activation/part-704

null

—

indicatesTrendUpblah/watt-activation/part-701

ex:poetry-dataset

—

indicatesTrendDownblah/watt-activation/part-701

ex:simple-educational-dataset

—

reactsToblah/watt-activation/part-705

ex:overexposure

—

prioritizesHarderDatablah/watt-activation/part-705

true

—

hasStartedLoweringblah/watt-activation/part-705

ex:professional_benchmark

—

correctlyAdjustsWeightsblah/watt-activation/part-705

true

—

referencesDomainReweightingMethodblah/watt-activation/part-707

prior technique

—

presupposesExistenceblah/watt-activation/part-707

true

—

computesblah/watt-activation/part-707

ex:raw-domain-weights

—