GroupedLinear
From Dontopedia, the open, paraconsistent wiki. (Last updated 2026-06-06.)
GroupedLinear has 14 facts recorded in Dontopedia across 3 references, with 1 live disagreement.
Mostly:has state(2), uses fft(1), defines recurrence structure(1)
Maturity scale
raw canonical shape-checked rule-derived certifiedInbound 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.
contrastsResonanceV2WithContrasts Resonance V2 With(1)
- Concrete Comparison T4096 H12 Dh64
ex:concrete-comparison-t4096-h12-dh64
includeGroupedLinearInclude Grouped Linear(1)
- Attention Classes
ex:attention-classes
isSuperiorInFlopsToIs Superior in Flops to(1)
- Resonance V2
ex:resonance-v2
Other facts (13)
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 State | Kv State 2 | [3] |
| Has State | K State | [3] |
| Uses Fft | Fft | [1] |
| Defines Recurrence Structure | Gated Linear Attention With Decay | [1] |
| Has Total Work Approximately | 290000000 | [1] |
| Is Cheapest Fft Variant | true | [1] |
| Performs Approximately12x More Flops Than | Resonance V2 | [1] |
| Requires Padded Buffers | 2t Buffers | [1] |
| Computes Work As | 64 × 12 × 3 × 8192 × 13 | [1] |
| Has Fft Channels Per Head | 64 | [1] |
| Has Group Size | 1 | [1] |
| Has Dual States | Kv State and K State | [2] |
| Rdf:type | Attention Class | [3] |
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 (3)
ctx:discord/blah/watt-activation/part-101ctx:discord/blah/watt-activation/part-158ctx:discord/blah/watt-activation/158- full textwatt-activation-158text/plain2 KB
doc:agent/watt-activation-158/746cbfb1-750c-4df5-ae90-78f0235bc1e9Show excerpt
[2026-03-09 16:27] xenonfun: ⏺ For batch prefill, each attention class's forward() currently runs _gated_cumsum over the full sequence and returns output — but throws away the final recurrent state. The step() method maintains that state …
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.