Dontopedia

swiglu

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

swiglu has 15 facts recorded in Dontopedia across 3 references, with 2 live disagreements.

15 facts·9 predicates·3 sources·2 in dispute

Mostly:involves operations(5), rdf:type(1), learnable param count(1)

Maturity scale raw canonical shape-checked rule-derived certified

Inbound mentions (1)

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.

activationCompetitorActivation Competitor(1)

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.

13 facts
PredicateValueRef
Involves OperationsMatmul Gate[3]
Involves OperationsSilu[3]
Involves OperationsMatmul Up[3]
Involves OperationsMul[3]
Involves OperationsMatmul Proj[3]
Rdf:typeActivation Function[1]
Learnable Param Count0[1]
NoteGated architecture[1]
Predicted LikelihoodMost Likely Winner[2]
Advantage TypeStructural Advantage[2]
Advantage ContextLow Step Counts[2]
Has Op Count5[3]
Has Parametric Difference50% more params[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.

typeblah/training-and-evals/23
ex:ActivationFunction
labelblah/training-and-evals/23
swiglu
learnableParamCountblah/training-and-evals/23
0
noteblah/training-and-evals/23
Gated architecture
predictedLikelihoodblah/training-and-evals/24
ex:most-likely-winner
advantageTypeblah/training-and-evals/24
ex:structural-advantage
advantageContextblah/training-and-evals/24
ex:low-step-counts
labelblah/training-and-evals/28
swiglu activation
involvesOperationsblah/training-and-evals/28
ex:matmul_gate
involvesOperationsblah/training-and-evals/28
ex:silu
involvesOperationsblah/training-and-evals/28
ex:matmul_up
involvesOperationsblah/training-and-evals/28
ex:mul
involvesOperationsblah/training-and-evals/28
ex:matmul_proj
hasOpCountblah/training-and-evals/28
5
hasParametricDifferenceblah/training-and-evals/28
50% more params

References (3)

3 references
  1. [1]234 facts
    ctx:discord/blah/training-and-evals/23
    • full texttraining-and-evals-23
      text/plain3 KBdoc:agent/training-and-evals-23/70bec41a-b6db-4999-9455-15d7176b4205
      Show excerpt
      [2026-02-25 14:27] omega [bot]: **Symbio Search Math Problem** Current run `super_chat_20260225140824_ee4k` has a config mismatch: ``` 50000 candidates × 3000 steps × 20 gens = 3,000,000,000 total search steps But only 50,000 total trainin
  2. [2]243 facts
    ctx:discord/blah/training-and-evals/24
    • full texttraining-and-evals-24
      text/plain2 KBdoc:agent/training-and-evals-24/b280a4e7-48a7-4bc2-9593-e4261e806744
      Show excerpt
      [2026-02-25 14:46] omega [bot]: **Selective Pressure — Three Mechanisms** **1. Evaluation Pressure** (per-candidate) Each candidate gets exactly 30 steps to prove itself. Records bestLoss, bestValLoss, fitnessScore. At 30 steps during warm
  3. [3]288 facts
    ctx:discord/blah/training-and-evals/28
    • full texttraining-and-evals-28
      text/plain3 KBdoc:agent/training-and-evals-28/d5fb7857-3e3e-4219-8c84-377b4bfc1def
      Show excerpt
      [2026-02-25 15:46] omega [bot]: **When Do Universal & KAN Spline Get Faster?** Short answer: **they don't get faster.** Their tok/s is fixed by their op count. But here's the full picture. **Op count per FFN forward pass (per layer × 8 la

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.