Comparison of Final Val PPL

as donto believed on

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

Comparison of Final Val PPL has 37 facts recorded in Dontopedia across 10 references, with 4 live disagreements.

37 facts·31 predicates·10 sources·4 in dispute

Mostly:rdf:type(3), compares model configuration(2), has comparison target(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.

`assertsComparison`Asserts Comparison(1)

Xenonfunex:xenonfun

`containsObservation`Contains Observation(1)

Log Entry 2026 03 09 03 00ex:log-entry-2026-03-09-03-00

`involvesComparison`Involves Comparison(1)

Experiment aex:experiment-a

Other facts (35)

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.

35 facts

Predicate	Value	Ref
Rdf:type	Performance Comparison	[1]
Rdf:type	Comparison	[6]
Rdf:type	Comparison	[10]
Compares Model Configuration	Spectral Lohe V2 37 4m	[4]
Compares Model Configuration	Spectral Gelu 92 9m	[4]
Has Comparison Target	vq_layer=1	[6]
Has Comparison Target	vq_layer=2	[6]
Compares Subject	Iteration 62500	[1]
Compares With	Iteration 62200	[1]
Sequence Length Change	4	[2]
Wall Time Multiplier	1.43	[2]
Implies	Quadratic Term Acceleration	[2]
Compares Subject1	Groupedlinear	[3]
Compares Subject2	Resonancev2	[3]
Metric	FLOPs	[3]
Ratio	~12x	[3]
Direction of Comparison	more	[3]
Parameter Ratio	40 percent	[4]
Quality Ratio	93 percent	[4]
Speed Ratio	74 percent faster	[4]
Inference About Wall Clock Time	would see more data	[4]
Inference About Gap Closing	likely close that gap further	[4]
Matches Pattern	Cuda Spectralgpt Result Pattern	[4]
Has Baseline	Fp32 Configuration	[5]
Has Comparison	Ternary 1 58 Bit Configuration	[5]
Percentage Difference	+6.9%	[5]
Training Budget	20000	[7]
Budget Unit	steps	[7]
Architecture	Dense Projection Logit Decoder	[8]
Has Better Performance	Rotor Block	[9]
Has Worse Performance	Helmholtz Block	[9]
In Context of	Ofdm 32	[9]
Compares	Metric Best Bpb	[10]
Has Result	better than predicted	[10]
Compares to	Predicted Bpb	[10]

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.

37 undated

—

typeblah/watt-activation/32

ex:PerformanceComparison

—

labelblah/watt-activation/32

Comparison: Not compiled vs Compiled

—

comparesSubjectblah/watt-activation/32

ex:iteration-62500

—

comparesWithblah/watt-activation/32

ex:iteration-62200

—

sequenceLengthChangeblah/watt-activation/57

—

wallTimeMultiplierblah/watt-activation/57

1.43

—

impliesblah/watt-activation/57

ex:quadratic-term-acceleration

—

comparesSubject1blah/watt-activation/101

ex:groupedlinear

—

comparesSubject2blah/watt-activation/101

ex:resonancev2

—

metricblah/watt-activation/101

FLOPs

—

ratioblah/watt-activation/101

~12x

—

directionOfComparisonblah/watt-activation/101

—

comparesModelConfigurationblah/watt-activation/123

ex:spectral-lohe-v2-37-4m

—

comparesModelConfigurationblah/watt-activation/123

ex:spectral-gelu-92-9m

—

parameterRatioblah/watt-activation/123

40 percent

—

qualityRatioblah/watt-activation/123

93 percent

—

speedRatioblah/watt-activation/123

74 percent faster

—

inferenceAboutWallClockTimeblah/watt-activation/123

would see more data

—

inferenceAboutGapClosingblah/watt-activation/123

likely close that gap further

—

matchesPatternblah/watt-activation/123

ex:cuda-spectralgpt-result-pattern

—

labelblah/watt-activation/276

Comparison of Final Val PPL

—

hasBaselineblah/watt-activation/276

ex:fp32-configuration

—

hasComparisonblah/watt-activation/276

ex:ternary-1-58-bit-configuration

—

percentageDifferenceblah/watt-activation/276

+6.9%

—

typeblah/watt-activation/278

ex:Comparison

—

hasComparisonTargetblah/watt-activation/278

vq_layer=1

—

hasComparisonTargetblah/watt-activation/278

vq_layer=2

—

trainingBudgetblah/watt-activation/324

20000

—

budgetUnitblah/watt-activation/324

steps

—

architectureblah/watt-activation/344

ex:dense-projection-logit-decoder

—

hasBetterPerformanceblah/watt-activation/393

ex:rotor-block

—

hasWorsePerformanceblah/watt-activation/393

ex:helmholtz-block

—

inContextOfblah/watt-activation/393

ex:ofdm-32

—

typeblah/watt-activation/460

ex:Comparison

—

comparesblah/watt-activation/460

ex:metric-best-bpb

—

hasResultblah/watt-activation/460

better than predicted

—

comparesToblah/watt-activation/460

ex:predicted-bpb

References (10)

10 references

[1]324 facts
ctx:discord/blah/watt-activation/32
- full textwatt-activation-32
  text/plain3 KBdoc:agent/watt-activation-32/8e8fa426-eb3a-4cc0-8c6a-550ba45f969f
  Show excerpt
  [2026-03-06 21:37] xenonfun: ====================================================================== Evaluation at iteration 20000 (~20% thru data) ====================================================================== Prompt: 'virtue is' Ge…
[2]573 facts
ctx:discord/blah/watt-activation/57
- full textwatt-activation-57
  text/plain2 KBdoc:agent/watt-activation-57/2be5a3f3-fa33-4f3c-809c-45db01bc0dea
  Show excerpt
  [2026-03-07 09:27] xenonfun: ``` Good question. Let me extrapolate from our existing data. Measured throughput (pairwise, batch=1): ┌─────────┬──────┬─────────┬───────┐ │ Seq Len │ it/s │ ms/iter │ tok/s │ ├─────────┼──────┼──────…
[3]1015 facts
ctx:discord/blah/watt-activation/101
- full textwatt-activation-101
  text/plain3 KBdoc:agent/watt-activation-101/6f90e9c2-2d77-484f-96fd-98a89c99440a
  Show excerpt
  [2026-03-08 18:08] xenonfun: ### Concrete comparison at T=4096, H=12, d_h=64: ResonanceV2 (cumsum, K=8 bands): Work: 12 × 8 × 4096 × 64 = 25M multiply-adds Kernel launches: ~K cumsums = 8 Metal kernels Memory: O(T · H · K · …
[4]1238 facts
ctx:discord/blah/watt-activation/123
- full textwatt-activation-123
  text/plain3 KBdoc:agent/watt-activation-123/2baaacb5-8c72-45ab-a16f-f5292a255289
  Show excerpt
  [2026-03-09 02:56] xenonfun: ``` val ppl=262.1 train ppl=223.9 321ms/step 12,779tok/s ──────────────────────────────────────────────────────────────────────── [kan+lohe_v2] params: 37.27M iter 50/1000 loss=8.6248 ppl= 5568.2 …
[5]2764 facts
ctx:discord/blah/watt-activation/276
- full textwatt-activation-276
  text/plain3 KBdoc:agent/watt-activation-276/53331a47-df58-4d3f-a691-96a8098e7b35
  Show excerpt
  [2026-03-13 21:31] xenonfun: ⏺ Results are in: ``` ┌───────────────┬───────────────┬────────────────────────────────┬───────┐ │ │ fp32 │ Ternary 1.58-bit │ Delta │ ├───────────────┼───────────────┼…
[6]2783 facts
ctx:discord/blah/watt-activation/278
- full textwatt-activation-278
  text/plain3 KBdoc:agent/watt-activation-278/f9f6e4c6-b1f8-402a-a4dc-d3b356140759
  Show excerpt
  [2026-03-13 22:26] xenonfun: ⏺ Exp 0 result: direction/gain split is safe. Key findings: ``` ┌─────────────┬─────────────┬───────────────────────────────────────┬────────────────────────────────────────────────┐ │ Metric │ Ea…
[7]3242 facts
ctx:discord/blah/watt-activation/324
- full textwatt-activation-324
  text/plain2 KBdoc:agent/watt-activation-324/5428b43d-e96d-4178-8924-6d7c75574aa4
  Show excerpt
  [2026-03-15 03:48] xenonfun: ## Design (files: Screenshot_2026-03-14_at_11.48.03_PM.png) [2026-03-15 03:49] xenonfun: (files: Screenshot_2026-03-14_at_11.49.32_PM.png) [2026-03-15 04:03] xenonfun: You went from 2.39 → 1.572 today. You cros…
[8]3441 fact
ctx:discord/blah/watt-activation/344
- full textwatt-activation-344
  text/plain3 KBdoc:agent/watt-activation-344/aa3dc64f-a03a-4cc7-a0ef-a9fda84ff8b8
  Show excerpt
  [2026-03-15 23:37] xenonfun: ``` ⏺ K is the Lohe coupling strength — it controls how strongly groups pull toward their mean field during sync. Currently frozen at init value 1/sqrt(G*H) ≈ 0.177. The critical coupling is K_c = H/(H-1) = 1.…
[9]3933 facts
ctx:discord/blah/watt-activation/393
- full textwatt-activation-393
  text/plain3 KBdoc:agent/watt-activation-393/d9206d9a-8fe3-4b8a-9bba-45db3179f660
  Show excerpt
  [2026-03-19 03:51] xenonfun: ⏺ Excellent results! Almost all done. Here's the encoding comparison with proper LR: ``` ┌─────────┬───────────┬───────────┬──────────┬───────┬───────┬────────┐ │ Config │ Encoding │ Block │ BPB @500 │…
[10]4604 facts
ctx:discord/blah/watt-activation/460
- full textwatt-activation-460
  text/plain3 KBdoc:agent/watt-activation-460/b1f5e9cd-e022-4479-b238-062663ee77df
  Show excerpt
  [2026-03-21 16:56] xenonfun: ⏺ Results on fineweb_edu_bytes.bin (623MB): ``` ┌────────────────┬──────────────────────────────────────┐ │ Metric │ Value │ ├────────────────┼───────────────────────…

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.

Comparison of Final Val PPL

Inbound mentions (3)

assertsComparisonAsserts Comparison(1)

containsObservationContains Observation(1)

involvesComparisonInvolves Comparison(1)

Other facts (35)

Timeline

References (10)

See also

Keep researching

`assertsComparison`Asserts Comparison(1)

`containsObservation`Contains Observation(1)

`involvesComparison`Involves Comparison(1)