Dontopedia

E23 Experiment

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

E23 Experiment has 21 facts recorded in Dontopedia across 2 references, with 1 live disagreement.

21 facts·19 predicates·2 sources·1 in dispute

Mostly:has goal(3), assumes ongoing(1), at matched steps with(1)

Maturity scale raw canonical shape-checked rule-derived certified

Inbound mentions (3)

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.

comparesCompares(1)

evaluatesGnormDecayAsHealthyEvaluates Gnorm Decay As Healthy(1)

monitorsGnormMonitors Gnorm(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 GoalGoal Validation[2]
Has GoalGoal Schedule[2]
Has GoalGoal Performance[2]
Assumes Ongoingtrue[1]
At Matched Steps WithE20 Experiment[1]
Behind by Bpb0.012[1]
Current Speed18.9K tok/s solo (no GPU contention)[1]
Depends on Lower Peak Lravoids wedge that hit E20 around step 1500–2000[1]
Estimated Remaining Time4 hr[1]
Has Gnorm Decay7.26 → 1.48 by step 600[1]
Has Lower Peak Lr ThanE20 Experiment[1]
Has Slightly Slower Start Becausepeak LR is half[1]
Has Slightly Worse Performance CurrentlyE20 Experiment[1]
Log Line Time~6 sec/log-line[1]
Potentially Superior toE20 Experiment[1]
Remaining Steps7400[1]
Rdf:typeExperiment Run[2]
Has ArchitectureE22 Nano Omni[2]
Has Optimization StackNew Optimization Stack[2]
Has Estimated Duration~7.5h[2]
Has Expected Step Time~0.9s/step[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.

assumesOngoingblah/watt-activation/part-691
true
atMatchedStepsWithblah/watt-activation/part-691
ex:e20-experiment
behindByBPBblah/watt-activation/part-691
0.012
currentSpeedblah/watt-activation/part-691
18.9K tok/s solo (no GPU contention)
dependsOnLowerPeakLRblah/watt-activation/part-691
avoids wedge that hit E20 around step 1500–2000
estimatedRemainingTimeblah/watt-activation/part-691
4 hr
hasGnormDecayblah/watt-activation/part-691
7.26 → 1.48 by step 600
hasLowerPeakLRThanblah/watt-activation/part-691
ex:e20-experiment
hasSlightlySlowerStartBecauseblah/watt-activation/part-691
peak LR is half
hasSlightlyWorsePerformanceCurrentlyblah/watt-activation/part-691
ex:e20-experiment
logLineTimeblah/watt-activation/part-691
~6 sec/log-line
potentiallySuperiorToblah/watt-activation/part-691
ex:e20-experiment
remainingStepsblah/watt-activation/part-691
7400
typeblah/watt-activation/696
ex:ExperimentRun
hasArchitectureblah/watt-activation/696
ex:e22-nano-omni
hasOptimizationStackblah/watt-activation/696
ex:new-optimization-stack
hasGoalblah/watt-activation/696
ex:goal-validation
hasGoalblah/watt-activation/696
ex:goal-schedule
hasGoalblah/watt-activation/696
ex:goal-performance
hasEstimatedDurationblah/watt-activation/696
~7.5h
hasExpectedStepTimeblah/watt-activation/696
~0.9s/step

References (2)

2 references
  1. [1]Part 69113 facts
    ctx:discord/blah/watt-activation/part-691
  2. [2]6968 facts
    ctx:discord/blah/watt-activation/696
    • full textwatt-activation-696
      text/plain2 KBdoc:agent/watt-activation-696/6bc363f6-e780-4242-9e13-32e8d01a4dd8
      Show excerpt
      [2026-05-01 02:47] xenonfun: It wants 150M or so so think the chatgpt-2 in 24hr is looking achievable. we do need to get more dataset here. [2026-05-01 02:50] lisamegawatts: mines downloading datasets and parsing now, added a 4096 option [2

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.