Step 3000

as donto believed on

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

Step 3000 has 43 facts recorded in Dontopedia across 6 references, with 3 live disagreements.

43 facts·37 predicates·6 sources·3 in dispute

Mostly:has loss(3), has bpb(3), has tokens per second(2)

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.

`includesStep`Includes Step(1)

Analytical Training Sessionex:analytical-training-session

`precedes`Precedes(1)

Step 2900ex:step-2900

`precedesStep`Precedes Step(1)

Analytical Training Sessionex:analytical-training-session

Other facts (43)

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.

43 facts

Predicate	Value	Ref
Has Loss	5.9616	[1]
Has Loss	5.094	[4]
Has Loss	5.094	[6]
Has Bpb	2.914	[3]
Has Bpb	7.349	[4]
Has Bpb	7.349	[6]
Has Tokens Per Second	719453	[3]
Has Tokens Per Second	65736	[4]
Has Gradient Norm	1256.97	[4]
Has Gradient Norm	1256.97	[6]
Has Time Ms	674	[1]
Has Lr	0.0000471	[1]
Has Tok Per S	12150	[1]
Has Ppl	388.2	[1]
Has Progress Percent	18.0%	[1]
Precedes	Step 3100	[1]
Has Eta Min	154	[1]
Time Per Step	382ms	[2]
Eta	60min	[2]
Lr	0.0000906	[2]
Loss	5.1453	[2]
Progress	24.1%	[2]
Ppl	171.6	[2]
Tok Per Sec	42866	[2]
Is Part of	Training Run 5000 Steps	[3]
Has Omega Component6	0.164	[4]
Has Cache Hit Percentage	99.4	[4]
Has Cache Size	500	[4]
Has Gamma	0	[4]
Has Omega Component1	0.189	[4]
Has Omega Component2	0.016	[4]
Has Omega Component3	0.154	[4]
Has Omega Component4	0.235	[4]
Has Omega Component5	0.049	[4]
Precedes Step	Step 6000	[4]
Has Step Number	3000	[5]
Part of Run	Run 1	[5]
Has Bpb Value	2.914	[5]
Has Throughput Tokens Per Second	719453	[5]
Rdf:type	Training Step	[6]
Step Index	3000	[6]
Has Token Speed	65736	[6]
Has Cache Hit Rate	99.4	[6]

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.

43 undated

—

hasTimeMsblah/watt-activation/part-153

674

—

hasLrblah/watt-activation/part-153

0.0000471

—

hasTokPerSblah/watt-activation/part-153

12150

—

hasLossblah/watt-activation/part-153

5.9616

—

hasPplblah/watt-activation/part-153

388.2

—

hasProgressPercentblah/watt-activation/part-153

18.0%

—

precedesblah/watt-activation/part-153

ex:step-3100

—

hasEtaMinblah/watt-activation/part-153

154

—

timePerStepblah/watt-activation/part-231

382ms

—

etablah/watt-activation/part-231

60min

—

lrblah/watt-activation/part-231

0.0000906

—

lossblah/watt-activation/part-231

5.1453

—

progressblah/watt-activation/part-231

24.1%

—

pplblah/watt-activation/part-231

171.6

—

tokPerSecblah/watt-activation/part-231

42866

—

hasTokensPerSecondblah/watt-activation/part-355

719453

—

isPartOfblah/watt-activation/part-355

ex:training-run-5000-steps

—

hasBpbblah/watt-activation/part-355

2.914

—

hasOmegaComponent6blah/watt-activation/part-482

0.164

—

hasBpbblah/watt-activation/part-482

7.349

—

hasCacheHitPercentageblah/watt-activation/part-482

99.4

—

hasCacheSizeblah/watt-activation/part-482

500

—

hasGammablah/watt-activation/part-482

—

hasGradientNormblah/watt-activation/part-482

1256.97

—

hasLossblah/watt-activation/part-482

5.094

—

hasOmegaComponent1blah/watt-activation/part-482

0.189

—

hasOmegaComponent2blah/watt-activation/part-482

0.016

—

hasOmegaComponent3blah/watt-activation/part-482

0.154

—

hasOmegaComponent4blah/watt-activation/part-482

0.235

—

hasOmegaComponent5blah/watt-activation/part-482

0.049

—

hasTokensPerSecondblah/watt-activation/part-482

65736

—

precedesStepblah/watt-activation/part-482

ex:step-6000

—

hasStepNumberblah/watt-activation/353

3000

—

partOfRunblah/watt-activation/353

ex:run-1

—

hasBpbValueblah/watt-activation/353

2.914

—

hasThroughputTokensPerSecondblah/watt-activation/353

719453

—

typeblah/watt-activation/480

ex:TrainingStep

—

stepIndexblah/watt-activation/480

3000

—

hasLossblah/watt-activation/480

5.094

—

hasBpbblah/watt-activation/480

7.349

—

hasTokenSpeedblah/watt-activation/480

65736

—

hasCacheHitRateblah/watt-activation/480

99.4

—

hasGradientNormblah/watt-activation/480

1256.97

References (6)

6 references

[1]Part 1538 facts
ctx:discord/blah/watt-activation/part-153
[2]Part 2317 facts
ctx:discord/blah/watt-activation/part-231
[3]Part 3553 facts
ctx:discord/blah/watt-activation/part-355
[4]Part 48214 facts
ctx:discord/blah/watt-activation/part-482
[5]3534 facts
ctx:discord/blah/watt-activation/353
- full textwatt-activation-353
  text/plain3 KBdoc:agent/watt-activation-353/cc7a24c1-66ae-472e-a74c-30bb70fe2a69
  Show excerpt
  [2026-03-17 09:19] xenonfun: ``` ============================================================ K4_cur10 K=4 curriculum=10% ============================================================ step 1000/5000 BPB=3.173 719,581 tok/s step 2…
[6]4807 facts
ctx:discord/blah/watt-activation/480
- full textwatt-activation-480
  text/plain2 KBdoc:agent/watt-activation-480/f2254c38-e451-4fa5-9d6b-499a16b05da3
  Show excerpt
  [2026-03-21 23:19] xenonfun: ``` === BPE-8K: cache growth 200→2000 === Loaded 20480128 bytes from /Users/ms/MS/HarmonicMLX/data/fineweb_edu_bpe8k.npy KickModel ANALYTICAL training: 30000 steps, seq_len=128, lr=0.0005, momentum=0.9…

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.

Step 3000

Inbound mentions (3)

includesStepIncludes Step(1)

precedesPrecedes(1)

precedesStepPrecedes Step(1)