Grouped1
From Dontopedia, the open, paraconsistent wiki. (Last updated 2026-06-06.)
Grouped1 has 25 facts recorded in Dontopedia across 3 references.
Mostly:has training throughput(1), is nearly five times faster than(1), is three point two times faster than(1)
Maturity scale
raw canonical shape-checked rule-derived certifiedInbound mentions (9)
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.
describesForwardThroughputDescribes Forward Throughput(1)
- Lisamegawatts Message 18 58
ex:lisamegawatts-message-18-58
describesScalingDescribes Scaling(1)
- Lisamegawatts Message 18 58
ex:lisamegawatts-message-18-58
describesTrainingThroughputDescribes Training Throughput(1)
- Lisamegawatts Message 18 58
ex:lisamegawatts-message-18-58
forFor(1)
- 22gb Peak
ex:22gb-peak
hasAttentionTypeHas Attention Type(1)
- Harmonicgpt
ex:harmonicgpt
hasWorkingBestHas Working Best(1)
- Harmonic Gpt
ex:harmonic-gpt
isSlowerByLargeMarginIs Slower by Large Margin(1)
- Anchor Kan
ex:anchor-kan
isSlowerInForwardIs Slower in Forward(1)
- Pytorch Mps
ex:pytorch-mps
isSlowerInTrainingIs Slower in Training(1)
- Pytorch Mps
ex:pytorch-mps
Other facts (25)
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 Training Throughput | 123000 | [1] |
| Is Nearly Five Times Faster Than | Pytorch Mps | [1] |
| Is Three Point Two Times Faster Than | Pytorch Mps Training | [1] |
| Has Latency Scaling | 6.4 | [1] |
| Exhibits Sub Linear Scaling | Latency Vs Tokens | [1] |
| Outperforms Spectral in Peak | Forward | [1] |
| Uses Cumsum | Implicit | [1] |
| Peaks at Forward Throughput | 905000 | [1] |
| Is Essentially | gated linear recurrence s_t = γ·s_{t-1} + k_t·v_t | [2] |
| Is Faster Than | Kan Quadratic | [2] |
| Is Handled Perfectly by | Gamma Scaled Cumsum | [2] |
| Is Winner | true | [2] |
| Would Be Faster Than | Anchor Kan | [2] |
| Has Description | element-wise gated linear attention | [2] |
| Has Ppl | 6.57 | [2] |
| Has Ppl Worseness Vs Spectral | 9% | [2] |
| Has Quality Adjusted Tokens Per Second | 18374 | [2] |
| Has Tokens Per Second at Temp256 | 516000 | [2] |
| Has Training Speed Multiplier Vs Kan | 10.7 | [2] |
| Is Element Wise Gated Linear Recurrence | true | [2] |
| Superior in Speed Quality Tradeoff | true | [2] |
| Rdf:type | Model or Configuration | [3] |
| Peak Memory Usage | 22 | [3] |
| Peak Memory Usage Unit | Gigabyte | [3] |
| Measured at Sequence Length | 4096 | [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-104ctx:discord/blah/watt-activation/part-103ctx:discord/blah/watt-activation/117- full textwatt-activation-117text/plain2 KB
doc:agent/watt-activation-117/93da7c3e-762b-4710-96fb-cdcbffaa6ad1Show excerpt
[2026-03-08 23:25] xenonfun: okay it still is using Adam finally pointed it at your optimizers: ``` ⏺ This is really interesting. There are 4 custom optimizers in harmonic-gpt/harmonic_gpt/optim/: 1. RotationalAdamW — Angular updates on …
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.