KAN

as donto believed on

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

KAN has 20 facts recorded in Dontopedia across 9 references, with 2 live disagreements.

20 facts·13 predicates·9 sources·2 in dispute

Mostly:rdf:type(4), has params(2), crashed due to(1)

Inbound mentions (4)

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.

`compareTo`Compare to(1)

Performance Claim L2048ex:performance-claim-l2048

`hasPplParityWith`Has Ppl Parity With(1)

Akan Modelex:akan-model

`intendsToInvestigate`Intends to Investigate(1)

Lisamegawattsex:lisamegawatts

`targetObject`Target Object(1)

Unpacking Phasesex:unpacking-phases

Other facts (17)

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.

17 facts

Predicate	Value	Ref
Rdf:type	Model Architecture	[5]
Rdf:type	Model	[6]
Rdf:type	Machine Learning Model	[8]
Rdf:type	Neural Network	[9]
Has Params	85,646,400	[2]
Has Params	49.5M	[3]
Crashed Due to	Diagnostics Block Unpack Phases	[1]
Will Be Re Run After Fix	null	[1]
Returns	None	[1]
Part of Top Performers	True	[2]
Return Value	None	[4]
Has Perplexity Value	91.3	[5]
Has Complexity	quadratic	[6]
Investigated for Capability	learn tokenize better	[8]
Combined With	Kuramoto Model	[8]
Tasked With Learning	similarities meaning	[8]
Performs Action	Shifting Byte Positions	[9]

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.

20 undated

—

crashedDueToblah/watt-activation/part-66

ex:diagnostics-block-unpack-phases

—

willBeReRunAfterFixblah/watt-activation/part-66

null

—

returnsblah/watt-activation/part-66

None

—

partOfTopPerformersblah/watt-activation/part-110

ex:true

—

hasParamsblah/watt-activation/part-110

85,646,400

—

hasParamsblah/watt-activation/part-647

49.5M

—

returnValueblah/watt-activation/66

None

—

typeblah/watt-activation/71

ex:ModelArchitecture

—

labelblah/watt-activation/71

KAN

—

hasPerplexityValueblah/watt-activation/71

91.3

—

typeblah/watt-activation/75

ex:Model

—

labelblah/watt-activation/75

KAN

—

hasComplexityblah/watt-activation/75

quadratic

—

labelblah/watt-activation/76

KAN

—

typeblah/watt-activation/177

ex:MachineLearningModel

—

investigatedForCapabilityblah/watt-activation/177

learn tokenize better

—

combinedWithblah/watt-activation/177

ex:Kuramoto-model

—

taskedWithLearningblah/watt-activation/177

similarities meaning

—

typeblah/watt-activation/416

ex:NeuralNetwork

—

performsActionblah/watt-activation/416

ex:shifting-byte-positions

References (9)

9 references

[1]Part 663 facts
ctx:discord/blah/watt-activation/part-66
[2]Part 1102 facts
ctx:discord/blah/watt-activation/part-110
[3]Part 6471 fact
ctx:discord/blah/watt-activation/part-647
[4]661 fact
ctx:discord/blah/watt-activation/66
- full textwatt-activation-66
  text/plain2 KBdoc:agent/watt-activation-66/b8a9d69a-c3a7-4d63-a909-00e2fb3be556
  Show excerpt
  [2026-03-07 15:56] lisamegawatts: there might be a way to use it for rope but we didn't find the right model [2026-03-07 15:57] lisamegawatts: it does embed positioning space, we may be looking in wrong location though [2026-03-07 15:57] xe…
[5]713 facts
ctx:discord/blah/watt-activation/71
- full textwatt-activation-71
  text/plain2 KBdoc:agent/watt-activation-71/82bde084-e631-42b5-99fe-0a0a8898ac2b
  Show excerpt
  [2026-03-07 17:14] xenonfun: ``` [E_mhkan_h5] iter 4000/10000 | avg_loss=4.7613 | PPL=116.9 | r=0.000 | E=0.000 | clusters=0 | 130 it/s (66.7K tok/s) │ │ [E_mhkan_h5] iter 6000/10000 | avg_loss=4.6…
[6]753 facts
ctx:discord/blah/watt-activation/75
- full textwatt-activation-75
  text/plain2 KBdoc:agent/watt-activation-75/24e73a49-4879-4ec1-9e4e-461a2fd9d66e
  Show excerpt
  [2026-03-07 18:23] xenonfun: Old per-anchor-loop benchmark complete. Key observations: 1. PPL parity confirmed — aKAN matches KAN within 0.3 PPL at every sequence length 2. Anchor entropy: H=3.46, eff_anc=31.8/32 — at random init / 2K …
[7]761 fact
ctx:discord/blah/watt-activation/76
- full textwatt-activation-76
  text/plain3 KBdoc:agent/watt-activation-76/961fc69a-4972-401d-be24-5f9157949baf
  Show excerpt
  [2026-03-07 18:31] xenonfun: ``` Excellent results. Full-sequence chunking (chunk_size=L) with the new formulation: ┌────────────────┬──────────────┬─────────────────────┬─────────────┬───────────┬────────┐ │ Config │ Old 5D to…
[8]1774 facts
ctx:discord/blah/watt-activation/177
- full textwatt-activation-177
  text/plain2 KBdoc:agent/watt-activation-177/1848e53e-9103-4e36-aa1e-3412420c8756
  Show excerpt
  [2026-03-09 23:15] xenonfun: ``` ⏺ Results are poor — actually worse than LoRA r16 (val ppl 287 vs 236). The core problem is becoming clear: it's the dataset quality, not the training method. NQ has multiple valid answers joined with " /…
[9]4162 facts
ctx:discord/blah/watt-activation/416
- full textwatt-activation-416
  text/plain3 KBdoc:agent/watt-activation-416/a471c17b-ef40-480f-b2f4-bbf6d25581f5
  Show excerpt
  [2026-03-19 21:25] xenonfun: ⏺ The spikes are in r and capacity — look at the oscillation: - Step 150: r=0.061, C=1.5b - Step 200: r=0.155, C=2.5b ← spike - Step 225: r=0.072, C=1.7b ← back down - Step 250: r=0.034, C=1.5b ← very l…

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.

KAN

Inbound mentions (4)

compareToCompare to(1)

hasPplParityWithHas Ppl Parity With(1)

intendsToInvestigateIntends to Investigate(1)

targetObjectTarget Object(1)