Dontopedia

Training Step 50

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

Training Step 50 has 52 facts recorded in Dontopedia across 8 references, with 5 live disagreements.

52 facts·40 predicates·8 sources·5 in dispute

Mostly:rdf:type(4), has loss(3), has lr(3)

Maturity scale raw canonical shape-checked rule-derived certified

Other facts (52)

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.

52 facts
PredicateValueRef
Rdf:typeTraining Step[5]
Rdf:typeTraining Step[6]
Rdf:typeTraining Step[7]
Rdf:typeTraining Step[8]
Has Loss4.9241[1]
Has Loss2.8955[3]
Has Loss4.9241[6]
Has Lr1.00e-04[1]
Has Lr0.0015[2]
Has Lr0.000204[3]
Has Bpb6.706[2]
Has Bpb4.177[3]
Has Bpb4.737[4]
Has Eta Minutes13[1]
Has Eta Minutes7[2]
Has Tokens Per Second77207[2]
Has Tokens Per Second10709[8]
Step Number50[7]
Step Number50[8]
Has Tok Per Second94840[1]
Has Step Time Ms1382[1]
Progress Percentage8.0%[1]
Occurs After StepTraining Step 25[1]
Has Ppl137.6[1]
Is Step Number50[1]
Has Temp0.89[2]
Has R0.795[2]
Has Progress Percent2.5[2]
Has Gnorm2.2132[3]
Has Tok Per S10709[3]
Has Step Number50[6]
Has Progress Percentage8[6]
Has Perplexity137.6[6]
Has Learning Rate at Step0.0001[6]
Has Step Duration1382[6]
Has Throughput94840[6]
Has Estimated Time Remaining13min[6]
Loss Value4.2575[7]
Bpb Value6.142[7]
R Value0.18[7]
Snr Value-14.8[7]
C Value2.7[7]
DC Value0.03[7]
Temperature Value0.57[7]
Learning Rate Value0.005[7]
Throughput Tokens Per Second21420[7]
Eta Minutes127[7]
Has Learning Rate Value0.005[7]
Has Loss Value2.8955[8]
Has Bpb Value4.177[8]
Has Gradient Norm2.2132[8]
Has Learning Rate0.000204[8]

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.

52 undated
—
hasEtaMinutesblah/watt-activation/part-268
13
—
hasTokPerSecondblah/watt-activation/part-268
94840
—
hasStepTimeMsblah/watt-activation/part-268
1382
—
progressPercentageblah/watt-activation/part-268
8.0%
—
occursAfterStepblah/watt-activation/part-268
ex:training-step-25
—
hasPplblah/watt-activation/part-268
137.6
—
hasLossblah/watt-activation/part-268
4.9241
—
isStepNumberblah/watt-activation/part-268
50
—
hasLrblah/watt-activation/part-268
1.00e-04
—
hasLrblah/watt-activation/part-386
0.0015
—
hasTempblah/watt-activation/part-386
0.89
—
hasBpbblah/watt-activation/part-386
6.706
—
hasRblah/watt-activation/part-386
0.795
—
hasProgressPercentblah/watt-activation/part-386
2.5
—
hasEtaMinutesblah/watt-activation/part-386
7
—
hasTokensPerSecondblah/watt-activation/part-386
77207
—
hasLossblah/watt-activation/part-658
2.8955
—
hasBpbblah/watt-activation/part-658
4.177
—
hasGnormblah/watt-activation/part-658
2.2132
—
hasTokPerSblah/watt-activation/part-658
10709
—
hasLrblah/watt-activation/part-658
0.000204
—
hasBpbblah/watt-activation/part-699
4.737
—
typeblah/watt-activation/229
ex:TrainingStep
—
typeblah/watt-activation/266
ex:TrainingStep
—
hasStepNumberblah/watt-activation/266
50
—
hasProgressPercentageblah/watt-activation/266
8
—
hasLossblah/watt-activation/266
4.9241
—
hasPerplexityblah/watt-activation/266
137.6
—
hasLearningRateAtStepblah/watt-activation/266
0.0001
—
hasStepDurationblah/watt-activation/266
1382
—
hasThroughputblah/watt-activation/266
94840
—
hasEstimatedTimeRemainingblah/watt-activation/266
13min
—
typeblah/watt-activation/427
ex:TrainingStep
—
stepNumberblah/watt-activation/427
50
—
lossValueblah/watt-activation/427
4.2575
—
bpbValueblah/watt-activation/427
6.142
—
rValueblah/watt-activation/427
0.18
—
snrValueblah/watt-activation/427
-14.8
—
cValueblah/watt-activation/427
2.7
—
dcValueblah/watt-activation/427
0.03
—
temperatureValueblah/watt-activation/427
0.57
—
learningRateValueblah/watt-activation/427
0.005
—
throughputTokensPerSecondblah/watt-activation/427
21420
—
etaMinutesblah/watt-activation/427
127
—
hasLearningRateValueblah/watt-activation/427
0.005
—
typeblah/watt-activation/655
ex:TrainingStep
—
stepNumberblah/watt-activation/655
50
—
hasLossValueblah/watt-activation/655
2.8955
—
hasBpbValueblah/watt-activation/655
4.177
—
hasGradientNormblah/watt-activation/655
2.2132
—
hasLearningRateblah/watt-activation/655
0.000204
—
hasTokensPerSecondblah/watt-activation/655
10709

References (8)

8 references
  1. [1]Part 2689 facts
    ctx:discord/blah/watt-activation/part-268
  2. [2]Part 3867 facts
    ctx:discord/blah/watt-activation/part-386
  3. [3]Part 6585 facts
    ctx:discord/blah/watt-activation/part-658
  4. [4]Part 6991 fact
    ctx:discord/blah/watt-activation/part-699
  5. [5]2291 fact
    ctx:discord/blah/watt-activation/229
  6. [6]2669 facts
    ctx:discord/blah/watt-activation/266
    • full textwatt-activation-266
      text/plain2 KBdoc:agent/watt-activation-266/0dd3318c-38a8-4ab0-8b7a-743748e72c54
      Show excerpt
      [2026-03-13 07:20] xenonfun: • Ran it. Long-prompt test (context_patches=128000, prompt = 1,024,000 bytes, generated 64 patches, compiled cached decode): - prompt bytes: 1,024,000 - generated patches: 64 - total elapsed: 231.8s …
  7. [7]42713 facts
    ctx:discord/blah/watt-activation/427
    • full textwatt-activation-427
      text/plain3 KBdoc:agent/watt-activation-427/40f37dca-9e0b-4399-abb4-7ccf7644f572
      Show excerpt
      [2026-03-20 02:39] xenonfun: ⏺ Pushed. The signing warning is cosmetic (SSH agent issue, push succeeded). Summary of what shipped: 1. Phase coupling (DYN-H4) — now the default in all 3 Lohe sync paths. Tangent magnitude IS the couplin…
  8. [8]6557 facts
    ctx:discord/blah/watt-activation/655
    • full textwatt-activation-655
      text/plain2 KBdoc:agent/watt-activation-655/184fbb55-c5df-4c57-ba7d-a525dd418a7a
      Show excerpt
      [2026-04-19 09:46] xenonfun: loaded 24999946 bytes from /Users/ms/MS/HarmonicRust/data/curriculum/domain_tinystories.bin step 0 loss=6.1092 bpb=8.814 gnorm=10.6885 lr=4.00e-6 9259 tok/s step 50 loss=2.8955 bpb=4.177 gnorm=2.2132…

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.

↑