Linear KAN
From Dontopedia, the open, paraconsistent wiki. (Last updated 2026-06-06.)
Linear KAN has 25 facts recorded in Dontopedia across 4 references, with 1 live disagreement.
Mostly:has throughput(2), has issue(2), has bpb(2)
Maturity scale
raw canonical shape-checked rule-derived certifiedInbound mentions (18)
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.
hasThroughputMultiplierOverHas Throughput Multiplier Over(2)
- Grouped Linear Kan
ex:grouped-linear-kan - Lohe Control
ex:lohe-control
isFasterThanIs Faster Than(2)
- Grouped Linear Kan
ex:grouped-linear-kan - Lohe Control
ex:lohe-control
bpbTiedWithBpb Tied With(1)
- Grouped Linear Kan
ex:grouped-linear-kan
comparesModelsCompares Models(1)
- Wave Native Lm Bakeoff
ex:wave-native-lm-bakeoff
concludesBestBPBConcludes Best Bpb(1)
- Lisamegawatts
ex:lisamegawatts
hasBPBAdvantageOverHas Bpb Advantage Over(1)
- Lohe Control
ex:lohe-control
hasNearIdenticalBPBToHas Near Identical Bpb to(1)
- Grouped Linear Kan
ex:grouped-linear-kan
is30PercentSmallerThanIs30 Percent Smaller Than(1)
- Lohe Control
ex:lohe-control
isCheaperThanIs Cheaper Than(1)
- Quadratic Kan
ex:quadratic-kan
isSignificantlyFasterAndMoreParamEfficientThanIs Significantly Faster and More Param Efficient Than(1)
- Lohe Control
ex:lohe-control
isSlightlyWorseOnBPBThanIs Slightly Worse on Bpb Than(1)
- Lohe Control
ex:lohe-control
outperformsOnSpeedOutperforms on Speed(1)
- Lohe Attention
ex:lohe-attention
presupposesExistenceOfPresupposes Existence of(1)
- Wave Native Lm Bakeoff
ex:wave-native-lm-bakeoff
recommendsForResearchBestPerformanceRecommends for Research Best Performance(1)
- Lisamegawatts
ex:lisamegawatts
tiesWithTies With(1)
- Grouped Linear Kan
ex:grouped-linear-kan
tradesSpeedForPerformanceTrades Speed for Performance(1)
- Lohe Attention
ex:lohe-attention
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.
| Predicate | Value | Ref |
|---|---|---|
| Has Throughput | 4.2 | [2] |
| Has Throughput | 4.2 | [4] |
| Has Issue | Full Dh Times Dh Outer Product Fft | [2] |
| Has Issue | hardware-unfriendly FFT on RTX 3060 | [4] |
| Has Bpb | 3.492 | [2] |
| Has Bpb | 3.492 | [4] |
| Is Linear in T | true | [1] |
| Uses Full Outer Product | Full Outer Product | [1] |
| Has Dh Squared Channels | 4096 | [1] |
| Has Enormous Constant Factor | true | [1] |
| Is Absurdly Expensive | true | [1] |
| Has Significant Computational Cost | null | [2] |
| Is Best on | Bpb | [2] |
| Has Params | 493000 | [2] |
| Achieves Better Performance | lower BPB | [3] |
| Has More Params Percentage | 44 | [3] |
| Preferred for Research | true | [3] |
| Suitable for Training Tolerance | slower ok | [3] |
| Trades Performance for Speed | Lohe Attention | [3] |
| Uses Fft Based Approach | true | [3] |
| Uses More Parameters Than | Lohe Attention | [3] |
| Is Criticized for | computational expense | [3] |
| Is Significantly Slower | true | [3] |
| Has Param Count | 493000 | [4] |
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.
References (4)
ctx:discord/blah/watt-activation/part-101ctx:discord/blah/watt-activation/part-648ctx:discord/blah/watt-activation/part-643ctx:discord/blah/watt-activation/645- full textwatt-activation-645text/plain3 KB
doc:agent/watt-activation-645/5c424827-0967-4bac-bbcc-cb1b4a153e35Show excerpt
[2026-04-16 02:39] lisamegawatts: oh nmd [2026-04-16 02:39] lisamegawatts: its all good [2026-04-16 02:40] lisamegawatts: i thought it still had the books stuff but it doesn't and you can filter it, so that should be plenty for 50m chinchil…
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.