Dontopedia

Model Training Run 1

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

Model Training Run 1 has 24 facts recorded in Dontopedia across 1 reference, with 3 live disagreements.

24 facts·21 predicates·1 sources·3 in dispute

Mostly:predicted bpb at step(2), prediction target step(2), prediction qualifier(2)

Maturity scale raw canonical shape-checked rule-derived certified

Other facts (24)

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.

24 facts
PredicateValueRef
Predicted Bpb at Step2[1]
Predicted Bpb at Step1.8[1]
Prediction Target Step5000[1]
Prediction Target Step10000[1]
Prediction Qualifierapproximately[1]
Prediction Qualifierupper-bound-threshold[1]
Reported Bpb at Step2.35[1]
Starting Bpb8.37[1]
Reached Step1800[1]
Has Convergence Statusclean convergence[1]
Has Stability Statusno instability[1]
Checkpoint Planned at Step2000[1]
Has Throughput7200[1]
Throughput Unittokens per second[1]
Has Throughput Qualifierapproximately[1]
Uses Batch Size16[1]
Uses OptionOption B[1]
Uses OptimizerRotational Adamw[1]
Has Convergence RateCurrent Rate[1]
Has Corpus Size35[1]
Corpus Size Unitmegabytes[1]
Corpus Size Qualifiersmall[1]
Overfitting Start Prediction8000[1]
Overfitting Prediction Qualifierapproximately[1]

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.

reportedBpbAtStepblah/watt-activation/642
2.35
startingBpbblah/watt-activation/642
8.37
reachedStepblah/watt-activation/642
1800
hasConvergenceStatusblah/watt-activation/642
clean convergence
hasStabilityStatusblah/watt-activation/642
no instability
checkpointPlannedAtStepblah/watt-activation/642
2000
hasThroughputblah/watt-activation/642
7200
throughputUnitblah/watt-activation/642
tokens per second
hasThroughputQualifierblah/watt-activation/642
approximately
usesBatchSizeblah/watt-activation/642
16
usesOptionblah/watt-activation/642
ex:option-b
usesOptimizerblah/watt-activation/642
ex:rotational-adamw
hasConvergenceRateblah/watt-activation/642
ex:current-rate
predictedBpbAtStepblah/watt-activation/642
2
predictionTargetStepblah/watt-activation/642
5000
predictionQualifierblah/watt-activation/642
approximately
predictedBpbAtStepblah/watt-activation/642
1.8
predictionTargetStepblah/watt-activation/642
10000
predictionQualifierblah/watt-activation/642
upper-bound-threshold
hasCorpusSizeblah/watt-activation/642
35
corpusSizeUnitblah/watt-activation/642
megabytes
corpusSizeQualifierblah/watt-activation/642
small
overfittingStartPredictionblah/watt-activation/642
8000
overfittingPredictionQualifierblah/watt-activation/642
approximately

References (1)

1 references
  1. [1]64224 facts
    ctx:discord/blah/watt-activation/642
    • full textwatt-activation-642
      text/plain3 KBdoc:agent/watt-activation-642/3d74a9d0-e733-41fb-ac2c-53a6f761291f
      Show excerpt
      [2026-04-16 02:04] xenonfun: ``` BPB 8.37 → 2.35 in 1,800 steps. Clean convergence, no instability. The first checkpoint at step 2,000 is about to save. Throughput steady at ~7.2K tok/s at batch=16 with Option B + RotationalAdamW. For

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.