Dontopedia
Explore

Unified Rotor Gpu 25m

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

Unified Rotor Gpu 25m has 75 facts recorded in Dontopedia across 2 references, with 4 live disagreements.

75 facts·63 predicates·2 sources·4 in dispute

Mostly:has best bpb at step(6), has bpb at step(6), has parameter count(2)

Maturity scale raw canonical shape-checked rule-derived certified

Has Parameter Countin disputehasParameterCount

  • 19000000[1]sourceall time · 679
  • 19[2]all time · Part 682

Crossed Bpb Threshold at Approx Stepin disputecrossedBpbThresholdAtApproxStep

  • 2000[2]all time · Part 682
  • 1500[2]all time · Part 682

Has Best Bpb at Stepin disputehasBestBpbAtStep

  • 1.71[1]sourceall time · 679
  • 1.61[1]sourceall time · 679
  • 1.65[1]sourceall time · 679
  • 1.56[1]sourceall time · 679
  • 1.86[1]sourceall time · 679
  • 2.2[1]sourceall time · 679

Has Bpb at Stepin disputehasBpbAtStep

  • 1.74[1]sourceall time · 679
  • 1.8[1]sourceall time · 679
  • 1.94[1]sourceall time · 679
  • 2.28[1]sourceall time · 679
  • 1.91[1]sourceall time · 679
  • 1.72[1]sourceall time · 679

Approaches Bpb ofapproachesBpbOf

Assumes Trajectory ThroughassumesTrajectoryThrough

Crossed Bpb ThresholdcrossedBpbThreshold

  • 1.89[2]all time · Part 682

Crossed Bpb Threshold of Lohe Delta E BindingcrossedBpbThresholdOfLoheDeltaEBinding

  • 2.1[2]all time · Part 682

Current Training StepcurrentTrainingStep

  • 7220[2]all time · Part 682

Has Best Bpb So FarhasBestBpbSoFar

  • 1.56[2]all time · Part 682

Has Best Bpb Value at Step1000hasBestBpbValueAtStep1000

  • 2.2[2]all time · Part 682

Has Best Bpb Value at Step2000hasBestBpbValueAtStep2000

  • 1.86[2]all time · Part 682

Inbound mentions (9)

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.

isEasyDoReMiCategoryIs Easy Do Re Mi Category(2)

isHardestDoReMiCategoryIs Hardest Do Re Mi Category(2)

comparesRunsCompares Runs(1)

givesOutlookGives Outlook(1)

improvesOverTimeSlowerImproves Over Time Slower(1)

storesDataForStores Data for(1)

subjectOfForecastSubject of Forecast(1)

Other facts (51)

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.

51 facts
PredicateValueRef
Has Best Bpb Value at Step30001.71[2]
Has Best Bpb Value at Step50001.65[2]
Has Best Bpb Value at Step70001.61[2]
Has Best Bpb Value at Step72001.56[2]
Has Best Bpb Value at Step72201.56[2]
Has Bpb Value at Step10002.28[2]
Has Bpb Value at Step20001.94[2]
Has Bpb Value at Step30001.91[2]
Has Bpb Value at Step50001.72[2]
Has Bpb Value at Step70001.8[2]
Has Bpb Value at Step72201.74[2]
Has Current Best Bpb Distance to Wavenative0.22[2]
Has Do Re Mi Weight Dense Science0.21[2]
Has Do Re Mi Weight Edu0.19[2]
Has Do Re Mi Weight Narrative0.06[2]
Has Do Re Mi Weight Tinystories0.06[2]
Has Remaining Steps to Total12800[2]
Has Throughput Tok Per Sec23400[2]
Is Ahead by Bpb1.2[2]
Is Ahead of Prior Run at Same StepBest Comparable Prior Run[2]
Is Slower Per Token by Factor2.7[2]
Predicted Best Bpb at Step200001.35-1.45[2]
Predicted Better ThanLohe Delta Variants[2]
Predicted Competitive WithWavenative 25m Production[2]
Represents Net WinWall Clock Time[2]
Uses MethodDoremi[2]
Crosses Threshold Earlier ThanLohe Delta a[2]
Expected to Be CompetitiveWavenative 25m Production[2]
Implies Superior Learning RatePrior Runs[2]
Improves Over TimeBpb Decrease[2]
Predicted to Be Much Better ThanLohe Delta Variants[2]
Presupposes Prior Runs ExistComparable Prior Runs[2]
Has Schedulecosine decay[1]
Advantageous inWall Clock Time[1]
Slowness Factor2.7[1]
Slower ThanLohe Delta 25m Replication[1]
Has Throughput23400[1]
Approaches Value1.392[1]
Has ComponentUnified Block[1]
Target IsWavenative 25m Production[1]
Steps Remaining12800[1]
Distance From Target Bpb0.22[1]
Crossed Bpb of2.10 at ~1500[1]
Crossed Bpb at Step1.89 at 2000[1]
OutperformsLohe Delta 25m Replication[1]
Outperforms by Margin~1.2 BPB[1]
Has MetricBpb[1]
Current Step7220[1]
Uses DataDoremi Dataset[1]
Rdfs:labelunified_rotor_gpu_25m[1]
Teleologically Aims for Low BpbTraining Objective[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.

advantageousInblah/watt-activation/679
ex:wall-clock-time
approachesBpbOfblah/watt-activation/part-682
ex:kura-pp-warm-restart-wavenative
approachesValueblah/watt-activation/679
1.392
assumesTrajectoryThroughblah/watt-activation/part-682
ex:cosine-decay
crossedBpbAtStepblah/watt-activation/679
1.89 at 2000
crossedBpbOfblah/watt-activation/679
2.10 at ~1500
crossedBpbThresholdblah/watt-activation/part-682
1.89
crossedBpbThresholdAtApproxStepblah/watt-activation/part-682
2000
crossedBpbThresholdAtApproxStepblah/watt-activation/part-682
1500
crossedBpbThresholdOfLoheDeltaEBindingblah/watt-activation/part-682
2.1
crossesThresholdEarlierThanblah/watt-activation/part-682
ex:lohe-delta-a
currentStepblah/watt-activation/679
7220
currentTrainingStepblah/watt-activation/part-682
7220
distanceFromTargetBpbblah/watt-activation/679
0.22
expectedToBeCompetitiveblah/watt-activation/part-682
ex:wavenative-25m-production
hasBestBpbAtStepblah/watt-activation/679
1.71
hasBestBpbAtStepblah/watt-activation/679
1.61
hasBestBpbAtStepblah/watt-activation/679
1.65
hasBestBpbAtStepblah/watt-activation/679
1.56
hasBestBpbAtStepblah/watt-activation/679
1.86
hasBestBpbAtStepblah/watt-activation/679
2.2
hasBestBpbSoFarblah/watt-activation/part-682
1.56
hasBestBpbValueAtStep1000blah/watt-activation/part-682
2.2
hasBestBpbValueAtStep2000blah/watt-activation/part-682
1.86
hasBestBpbValueAtStep3000blah/watt-activation/part-682
1.71
hasBestBpbValueAtStep5000blah/watt-activation/part-682
1.65
hasBestBpbValueAtStep7000blah/watt-activation/part-682
1.61
hasBestBpbValueAtStep7200blah/watt-activation/part-682
1.56
hasBestBpbValueAtStep7220blah/watt-activation/part-682
1.56
hasBpbAtStepblah/watt-activation/679
1.74
hasBpbAtStepblah/watt-activation/679
1.8
hasBpbAtStepblah/watt-activation/679
1.94
hasBpbAtStepblah/watt-activation/679
2.28
hasBpbAtStepblah/watt-activation/679
1.91
hasBpbAtStepblah/watt-activation/679
1.72
hasBpbValueAtStep1000blah/watt-activation/part-682
2.28
hasBpbValueAtStep2000blah/watt-activation/part-682
1.94
hasBpbValueAtStep3000blah/watt-activation/part-682
1.91
hasBpbValueAtStep5000blah/watt-activation/part-682
1.72
hasBpbValueAtStep7000blah/watt-activation/part-682
1.8
hasBpbValueAtStep7220blah/watt-activation/part-682
1.74
hasComponentblah/watt-activation/679
ex:unified-block
hasCurrentBestBpbDistanceToWavenativeblah/watt-activation/part-682
0.22
hasDoReMiWeightDenseScienceblah/watt-activation/part-682
0.21
hasDoReMiWeightEdublah/watt-activation/part-682
0.19
hasDoReMiWeightNarrativeblah/watt-activation/part-682
0.06
hasDoReMiWeightTinystoriesblah/watt-activation/part-682
0.06
hasMetricblah/watt-activation/679
ex:bpb
hasParameterCountblah/watt-activation/679
19000000
hasParameterCountblah/watt-activation/part-682
19
hasRemainingStepsToTotalblah/watt-activation/part-682
12800
hasScheduleblah/watt-activation/679
cosine decay
hasThroughputblah/watt-activation/679
23400
hasThroughputTokPerSecblah/watt-activation/part-682
23400
impliesSuperiorLearningRateblah/watt-activation/part-682
ex:prior-runs
improvesOverTimeblah/watt-activation/part-682
ex:bpb-decrease
isAheadByBpbblah/watt-activation/part-682
1.2
isAheadOfPriorRunAtSameStepblah/watt-activation/part-682
ex:best-comparable-prior-run
isSlowerPerTokenByFactorblah/watt-activation/part-682
2.7
outperformsblah/watt-activation/679
ex:lohe-delta-25m-replication
outperformsByMarginblah/watt-activation/679
~1.2 BPB
predictedBestBpbAtStep20000blah/watt-activation/part-682
1.35-1.45
predictedBetterThanblah/watt-activation/part-682
ex:lohe-delta-variants
predictedCompetitiveWithblah/watt-activation/part-682
ex:wavenative-25m-production
predictedToBeMuchBetterThanblah/watt-activation/part-682
ex:lohe-delta-variants
presupposesPriorRunsExistblah/watt-activation/part-682
ex:comparable-prior-runs
labelblah/watt-activation/679
unified_rotor_gpu_25m
representsNetWinblah/watt-activation/part-682
ex:wall-clock-time
slowerThanblah/watt-activation/679
ex:lohe-delta-25m-replication
slownessFactorblah/watt-activation/679
2.7
stepsRemainingblah/watt-activation/679
12800
targetIsblah/watt-activation/679
ex:wavenative-25m-production
teleologicallyAimsForLowBpbblah/watt-activation/part-682
ex:training-objective
usesDatablah/watt-activation/679
ex:doremi-dataset
usesMethodblah/watt-activation/part-682
ex:doremi

References (2)

2 references
  1. customctx: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
  2. [2]Part 68244 facts
    customctx:discord/blah/watt-activation/part-682

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.